Team Defensive Metrics

Team Defensive Metrics

Up to now, football analytics has been a little lop-sided. One reason is that on-the-ball data has been readily available, but off-the-ball data has not, so the focus has been on attacking metrics, and defensive metrics have lagged behind. This balance will eventually be redressed by the adoption of tracking data, but this is quite complicated to analyse, and it is doubtful whether many clubs yet have the expertise to take full advantage of it. In the meantime however, the Stratagem company capture some defensive statistics which seem to provide useful insights. In this post I want to explore two of them.

The first statistic, which I call pIG, (Players In front of Goal) is the number of defensive players in direct line from the shooting player and the goal (including the goalie).  The second statistic is a somewhat more subjective variable called defensive pressure (here denoted dPress).  dPress ranges from 0 to 5. A dPress of 0 indicates no players in the vicinity of the shot, and no-one blocking the shot. A dPress of 5 indicates the most intense pressure where for example  a player is held while taking a shot, or tackled by several players at once, or crowded out while making a header.

This post takes a largely qualitative look at pIG and dPress to see what kinds of things they can tell us.  The data comes from 282 matches played in the EPL between the 23rd of August 2016 and the 19th of March 2017.

Defensive Metrics by Type of Chance and Chance Creation

Stratagem classify attempts on goal by type of chance and type of chance creation.  The table below shows the average pIG for the commonest types.  A Dangerous Moment is an occasion where there is an opportunity to shoot, but a shot is not always taken, for example a ball played across the 6-yard box which a striker just fails to make contact with.

ChanceChance Creation
Chance TypeAverage pIGAverage dPRESS Chance Creation TypeAverage pIGAverage dPRESS
Penalty1.00.1Penalty Earned 1.00.0
Dangerous Moment1.62.1Previous shot2.01.7
Open Play2.71.9Cross2.12.0
Direct Free-Kick3.21.4Open Play Pass2.71.8
Free Kick2.92.2
Corner3.72.5

This makes sense, but doesn’t really tell us much we might have otherwise guessed.  But splitting the data by team reveals some interesting differences.  For instance, Figure 1  shows how different EPL teams defend headed and kicked corners (and yes the sample sizes are smaller than I would like, but differences between the teams are statistically significant.)

Fig. 1. Defending Corners

 

Defending Dispositions by Minutes Played

Figures 2a and 2b show how pIG and dPress change over the course of a match. (From here on, penalties and direct free-kicks are excluded. )

Fig. 2a. pIG by Minutes Played: Defending Team  

Fig. 2b. Defensive Pressure by Minutes Played: Defending Team

                                              

A number of pIG patterns stand out. Teams like Leicester, Everton, Sunderland, Stoke  and West Brom have relatively flat pIG profiles while Burnley is unique in having a higher number of defenders in front of goal than all the other teams at every stage in the match. Chelsea and Watford start cautiously, maitaining a high pIG early in the match; their pIG then drops as the first half progresses, and stays constant throughout the second half; Palace show the opposite behaviour, keeping more defenders in front of goal in the first half, and dropping the number as the second half progresses. Finally Liverpool seem unique in keeping few defenders in front of goal in the first half, while increasing protection as the match progresses.

dPress statistics seem less prone to variation. Most teams have flat dPress profiles, although Arsenal, Bournemouth and Burnley apply slightly less defensive pressure as the match progresses, and Tottenham slightly more.  Liverpool is unique in applying substantially more defensive pressure in the latter stages of the match, while Southampton show the opposite pattern, applying less pressure as the match progresses.

We can also view pIg and dPress from the perspective of the attacking team. Figures 3a and 3b show respectively the number of opposition defenders each team encounters, and the defensive pressure they face, as the match progresses:

Fig. 3a. pIG by Minutes Played: Attacking Team 

                                               

Fig. 3b. Defensive Pressure by Minutes Played: Attacking Team

Several different pIG patterns are evident here. Arsenal, Chelsea, Middlesborough, Manchester City and Watford face a constant number of defenders in the first half, and a progressively reducing number in the second half. Everton, Swansea and Tottenham face progressively fewer defenders as the match progresses,  while Sunderland, West Brom and Stoke face progressively more.  Arsenal face more defensive pressure in the second half than in the first, while Sunderland and West Brom, and to a smaller extent Bournemouth face increasing pressure as the match progresses. Middlesborough on the other hand face less defensive pressure in the later stages.

Effect of Game State on Defensive Metrics

It is generally accepted that many match statistics are heavily influenced by the current scoreline or game state; for example teams that are behind in a match shoot more often and cross the ball more often. How does game state affect pIG and dPress?

Fig. 4. pIG and dPress by Game State

Figure 4 shows that pIG increases with game state, which is what we might expect; most teams throw men forward when they are behind, and so have fewer defenders in front of their own goal, but when they are in front and protecting a lead, they adopt a more defensive strategy. Teams who are ahead also put more pressure on their opponents’ shots, although they seem less inclined to apply that kind of pressure when they are 3 or more goals in front. Presumably most games are already won by this stage, and the winning team can afford to slacken off.

Figures 5a and 5b show how Game Sate affects the defending of individual teams.

Fig. 5a. pIG by Game State

Fig. 5b. Defensive Pressure by Game State

These figures show some interesting defensive styles. Manchester City for instance tend to station relatively few players in front of goal when facing a shot, irrespective of  game state; however they do apply more defensive pressure when ahead in the game. Liverpool defend quite differently when behind and when in front;  when behind they have fewer players in front of goal, and defend by applying considerable pressure to the shot; when leading they have rather more defenders protecting the goal, but apply substantially less pressure.  Burnley consistently rely on keeping defenders in front of the goal throughout, and are reluctant to apply much pressure to shots at all.

Defensive Metrics and Expected Goals

As far as I know, the kind of defensive metrics described here have not been included in Expected Goals models.  To see the relationship of the metrics to scoring I estimated the logistic regression below where the ith attempt is coded 1 for a goal and 0 otherwise.  (Attempts following set pieces were excluded.)

 

attempt_i = & \beta_0 + \beta_1 pIG_i + \beta_2 dPress_i + e_i

 

The unstandardized coefficients for pIG and dPress are both negative and significant( β1 = -0.74, p <.001, and β2 = -0.10, p < .01). This means that increasing the number of defenders in front of the goal, and increasing the pressure on the shooting player decreases the likelihood of scoring. It is certainly plausible that including this kind of metric in an xG model could improve the predictions, and this might be useful in assessing player goal-scoring performance.

Finally

This brief overview of  pIG and dPress suggests they are reasonably well-behaved statistics that satisfy some basic sanity checks, and I think they deserve further analysis. I’ll be writing a bit more about them soon…

 

This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform, in addition to StrataBet Premium Recommendations

2017-07-14T08:34:14+00:00 March 30th, 2017|Football Analytics, On the Pitch, Recent Posts|2 Comments

2 Comments

  1. Luis Pacheco May 16, 2017 at 4:39 pm - Reply

    Fantastic analysis! Would you post the R corde somewhere or odo you have a Github account

    • admin May 29, 2017 at 10:33 am - Reply

      Hi Luis

      Thanks for your comment. Sorry for the delay in responding, but I’ve had an operation and am just recovering. I’ll email you.

Leave A Comment