Ibson battles with former Loon Mo Saeid. Image credit to Daniel Mick - www.danielmickcreative.com

The Angle

The Value of Possession

by on 21 November 2017

As statistics begin to play a larger role in watching and analyzing soccer, the role of possession provides context for the action on the field. Dave Laidig argues that we can use possession data to understand how likely one is to score a goal.

Soccer is a dynamic team sport of constant movement over 90 plus minutes. And while scoring goals deservedly draws attention; most of the match is spent in coordinated runs and feints as teams try to advance the ball into dangerous spots near goal, or seek to deny opponents the same. In this effort, possession closer to the opponent’s goal obviously increases one’s chance of scoring. However, it is not easy to say how much gaining a few yards on the field is worth.

If we know the typical result for a touch in a specific area (such as the penalty area, center circle, or near the touchline), one of the things we can do is figure out how much more or less effective a particular team is at turning possession in one area of the pitch into a goal. One can create a high-level diagnostic tool for teams by comparing the team’s values to the typical result for the league. This contrast highlights areas having problems or doing well (for the team or opponents). This is something similar to a dashboard warning light: i.e., something is wrong here and it should be checked out. For example, if a team largely under performs the average in a particular part of the pitch, we can say that a position, player, or tactic is not using possession to create enough scoring opportunities.

Further, the likelihood of scoring from certain areas can inform individual actions as well. For example, not all completed passes are equal. With zone values, we can determine whether the ball moved from a low value area to a high value area. Adding together a large number of passes over a game or season, we can determine how much value a player is adding. And this has the added benefit of being independent of players further down the chain. In contrast, to get an assist, the pass recipient must score. And while expected assists (xA) and xGChain are not dependent on scoring, the pass recipient still must shoot to get credit under these stats. Being able to isolate a player’s value has its own utility and may be fairer for the players earlier in the chain of possession.

Before addressing the value of certain actions, we need to determine how likely a player is to score based on the field location. And because soccer involves chains of player actions, we’ll consider it in terms of possession. In other words, what are the chances of scoring at the end of a possession chain, given that the player is possessing the ball in a specified spot?

Defining a possession

Before diving into data, it is worth addressing what a “possession” is. For this endeavor, I wanted to analyze chains of offensive touches where a team controls the ball. I wanted “possession” to mean “deliberate action on offense” – or as close as I could get to it. For me, deliberate offensive actions include: shots, dribbles, and completed passes (allowing the assumption the passer intended to complete the pass).

I don’t consider every pass attempt part of a possession due to my experience in hand coding games. Specifically, sometimes emergency defending or aerial challenges show up as incomplete passes. And although reasonable people may differ, these actions do not seem to fit the “deliberate action on offense” criteria. Consequently, incomplete passes require an extra showing that the player had some time to prepare the next action to be counted as in possession.

For our purposes, passes after throw-ins, corners, and free kicks are presumed to allow such time. In addition, if the same player had a defensive action (tackle, interception) and the subsequent pass, we’ll assume the player had enough time on the ball for a deliberate pass and to establish possession.  If it’s not the same player, it’s assumed that the defense action broke up the play and another player made the clearance. And keep in mind that the definition may not match the eye test 100% of the time, but the rules are intended to be consistent and match our goal of deliberate action on offense more often than not.

In sum, we have the following definition of what it takes to establish a possession. Possessions begin at:

But once a possession is established, it is difficult to end the possession. I like being results oriented. Regardless of how the ball is advanced (fantastic dribbling, cutting pass, or punting it at the nearest opponent and having the deflection go out of bounds 20 yards away); all that matters is whether the team continues to control the ball.

Possessions end at:

In effect, nearly every offensive touch is in possession. Only about 5% of offensive touches are “not possession”, and these most likely are clearances (relief of defensive pressure) and aerial challenges before a team demonstrates control.

Calculating chances of scoring from a location

This possession definition was applied to all offensive and defensive actions in the 2017 MLS season (374 games). Then, the possession result was assigned every touch in the possession. If there was no shot at the end of the possession, every touch in that chain would be worth zero. If there was a shot, each touch in the chain would get assigned the end-result xG. (Brief refresher, expected goals or xG is a measure of shot quality and represents the likelihood of a goal – under similar circumstances as the shot.) In this process, each touch was assigned the probability of scoring for the same possession.

Once all of the possessions in each match were coded, and the possession result assigned; it’s possible to calculate the probability of scoring from each zone. The xG for all touches in each zone is summed up, and divided by the number of possession-touches in that zone. In other words, the resulting value represents the average xG resulting from a touch in possession in that particular zone.

We used zones that roughly align with markings on the pitch. For example, the 6-yard box is divided into three zones. The first zone is the width of the goal, and the other sizes reflect the remaining portions of the 6-yard box on either side of the goal. Similarly, for the ease of translating the values to what one can easily see, other zones have common boundaries; such as the width of the goal, edges of the penalty areas, midfield line, etc. The attacking third is divided into more zones because smaller changes were more meaningful than in the defending third.

The following chart shows the calculated xG per touch values for each zone from this analysis.

 

The values in the above chart are the average possession result (in xG) for an offensive touch in that zone. And as a spot check on the reasonableness of the values, we see that moving the ball closer to the opponent’s goal increases the chances of scoring, as would be expected. In addition, moving the ball centrally also increases the likelihood of scoring.

The center of the penalty area has a value of 0.190. This means possessing the ball in this spot yields, on average, a shot with 0.19 xG. For every five touches in this area, we would expect about 0.95 xG to result. We could say this this zone represents about a 19% chance of scoring.

This doesn’t mean that there necessarily will be one goal for every five touches; because there is always some randomness and players will go on runs and suffer cold streaks. But with a large enough sample (like a whole season), these values show the average result from a touch in possession. Similarly, we can demonstrate the relative value of possessing the ball in the various zones.

For another example, possessing the ball in the attacking side of the center line typically yields, on average, 0.018 xG. Here, it would take about 55 touches to get to the 1 xG mark. In relative terms, that makes the likelihood of scoring from this area about 11 times lower than the center of the penalty area.

Applying zone probabilities

Knowing the probability of scoring from different locations can serve as the foundation for other analyses. For one example, taking the ball from a non-dangerous spot and delivering right in front of the opponent’s goal obviously adds value. Using the probability grid and comparing the starting spot and the delivery location, we can quantify how much value was added from a single action.

Consider a quick counter attack from the back. The goalkeeper (in his or her own penalty area) punts the ball just across midfield; from a 0.008 xG zone to 0.018. This adds a small increase in the chances of scoring, about 0.010 xG.

A midfielder collects the ball, advances a bit and passes the ball to a winger in the left corner, about even with the penalty box. These actions advanced the ball from a 0.018 zone to 0.025, which increases the probability of scoring about 0.007 xG.

The winger looks up, and delivers a cross to a striker in the center of the penalty area; moving the ball from a 0.025 xG zone to 0.190 xG. This winger added 0.165 xG in value, and the striker had the ball in a high value location, where one could expect a goal about every five touches.

In this short possession chain, the players contributed the following amounts:

And these values exist regardless of whether the striker muffed the shot, lost his dribble, or scored a golazo – which allows better comparisons between players on Colorado to those on Toronto. Further, we can also apply the probability grid to turnovers. If the cross from the winger was collected by the opponent’s keeper instead; the action would have started from a 0.025 xG zone and ended up with 0.008 xG for the opponent (or negative 0.008 from the winger’s perspective). The result minus the starting value makes it -0.008 minus 0.025. Thus, the turnover results in -0.033 xG for the failed cross.

With a collected cross, the players contributed the following amounts:

In this outcome, we see that the keeper and midfielder values don’t change based on the possession result. But the winger now has a negative value, showing the cross from the side was a bit high risk, high reward.

These examples simplify the context a bit, but it highlights the basic utility of the probability grid. We can estimate the values of game actions based on league averages. As the example shows, the margins are pretty small individually. But when a team may have 400-600 touches per game, and 34 matches in a season, patterns of performance can show up. Zone values may not be a standalone metric, but it can serve as the measuring system to evaluate subsets of game action.


FiftyFive.One is now on Patreon. Do you like the independent coverage of soccer news from Minnesota and beyond that FiftyFive.One offers? Please consider becoming a patron.

Tags: , , , , , , , ,

  • Governor Squid

    I was told there would be no math.