Using Expected Goals to highlight defensive contributors
In last week’s blog, Mark Taylor introduced a much more inclusive alternative model to the now familiar shot based expected goals (xG) metric, using it to quantify the attacking contribution of creative passers. This week we’ll look at the often neglected area of defensive metrics and show how we can go beyond the mere counting stats that is indiscriminately used to characterise defensive excellence.
A non-shot expected goals model put’s a price on every possession enjoyed by a team throughout the course of a match, based on the typical outcome from a large historical database of previous such possessions.
A possession on the edge of your own penalty area has an extremely small likelihood of eventually culminating in a goal. But if the possession is owned on the edge of your opponent’s box, the chance of a goal resulting is much larger.
The advantages of such a model primarily derives from being able to separate the contribution made by individual players to all potentially successful match events, whether they result in an attempt on goal or not.
It is churlish at best to deny a player credit for a skilled pass that opens up a defence, but then the attack stalls before a chance can be attempted.
By moving away from a purely attempt based model, we can begin to evaluate the hither unseen contributions from players who have a less visible, but nevertheless integral roles to fulfil for their side.
Football is ultimately a team sport, and while the two contenders during the 2017/18 awards season, Salah and De Bruyne have been at the forefront of their team’s attacking output, there have been equally deserving, but more defensively orientated recipients in the recent past.
N’Golo Kante, for example and we’d like to be able to evaluate such performances using the same units of measurement, expected goals.
Let’s flesh out the methodology with a simple example.
Imagine player A has the ball in a central position just inside his or her opponent’s half of the field.
We’ll let the non-shot xG for Player A in possession of the ball at that spot on the field equal X.
If the player attempts to find a colleague on the edge of their opponent’s penalty area, the non-shot xG for the team will increase to Y, should the pass be successfully executed.
However, if a defensive player, we’ll call them Player D, manages to take possession of the intended pass, he or she may gain possession on the edge of their own box, along with a much smaller non-shot xG figure for their side.
Which we’ll call Z.
We’re left with a simple calculation that quantifies what both the attacking and defensive team could gain from this single pass amongst hundreds made in a single game, expressed in the extremely useful units of expected goals.
If the pass is successful, the attacking side, particularly the passer and the intended target, improve their side’s NS xG from X to Y.
If the defender takes possession in the region of the intended pass, they take away the value of their opponent’s possession prior to the pass, namely X and accrue a small amount of NS xG for themselves, in this case Z.
All we need to add is the likelihood that this pass will be completed or not from analysis of our historical database of passes and we can reach a numbers based conclusion for the value of the attacking pass or the defensive action and compare the actual records for passers and would be tacklers to evaluate their performances against a league standard.
Here’s a summary of the defensive contributions made by Liverpool players with a minimum of 900 minutes of Premier League action prior to last week’s match with Stoke.
We’ve taken the NS xG for the opponent making the pass and combined it with the NS xG from the perspective of the player after they force an incompletion.
The amount of NS xG “won” by each player is then compared to the overall record of Liverpool as a side during the 2017/18 Premier league and the percentage of time spent on the field by each player is also accounted for.
Opportunity, plus designated role within the team often drives each player’s statistics.
For example, Joe Gomez had played for 5.2% of the total time played by every Liverpool player prior to the game at home against Stoke and he was responsible for 5.3% of the NS xG “won” by his side by disrupting passes.
Player of the Year, Mo Salah has appeared in nearly 8% of Liverpool’s total minutes, but contributed just 2.3% of their tackle based NS xG recovery.
Although, it’s fair to say few are complaining about his lack of defensive involvement. Salah’s major primary role at Liverpool is as an attacking asset, Gomez has much more defensive duties.
Such numbers often highlight players who graft in the centre of the pitch, but rarely show up on the highlight reels that are concerned mainly with goals and scoring attempts.
Jordan Henderson, is a fine example.
His defensive contribution is nearly double the share of his time spent on the field, while Oxlade-Chamberlain, Can and laudably for an attacking player, Firmino also make sizeable relative contributions, based on their appearance time and primary roles.
Not only does this alternative approach highlight less heralded contributors, it also allows us to begin to quantify and adjust team quality based on the starting eleven selected on the day.