Does Footedness Really Matter?

Virgil Van Dijk’s assist against Bayern Munich
Laporte’s long pass against Newcastle United

Left Footed, Right Footed — Does it Make A Difference?

Football community has a divided opinion about this idea — should defenders be made to play on the natural side of the field based on their preferred foot?

Data:

We make use of the publicly available Wyscout’s spatiotemporal match events dataset, which logs events like passes, shots fouls etc. It stores details like event position coordinates, player who performed the event, timestamp etc. for each event for 2017–18 season. To obtain the player positional data in the starting lineup for each match, we relied on the formation data on fbref.com (on consensual basis). For this exercise, we use data from five leagues — English Premier League, Ligue 1, Serie A, La Liga and Bundesliga.

Assumptions:

Before we jumped onto implementation, we made the following assumptions which are relevant for our problem statement.

  1. We focus our study on left centre backs as they provide a good mix of left and right footed players (All right centre backs tend to be right footed).
  2. Since defenders usually tend to progress the ball by passing more than carrying, we have filtered out progressive passes from the data using Wyscout’s progressive pass definition . According to the definition, a pass is classified as progressive if it is at least 30m if starting and ending pass coordinates are within the team’s own half, at least 15m if starting and ending pass coordinates are in different halves and at least 10m if starting and ending pass coordinates are in the opponent’s half.
  3. We consider left centre backs in four defender formation and in three/five defender formation as different entities (L_CB and LCB respectively) to capture the impact of change in formation.
  4. We only consider the ones who have played at least 20 progressive passes at a particular position for a team to avoid including players with less progressive pass data.

Method:

Once we have our assumptions defined, we move ahead by working on the following pipeline:

  1. Firstly, we merged events data with player positional data as they come from different sources. Since we had inconsistencies in player and team names in both data sources, we merged these by creating a common reference vocabulary for player and team names, using accent handling and string matching. We had to manually intervene in a limited number of cases.
  2. We further adjusted the Wyscout’s pitch coordinates to the pitch size of 104 x 64 m. We also divided the pitch into eight regions as shown below in Fig 1. For our analysis on ball progression into attack, we concentrated on the four attacking regions.
  3. We then computed the desired metrics — preference and offensive value-added per progressive pass (OVPP) — for each region. For OVPP values, we relied on the training a VAEP framework using our data (as explained in the next section)
Fig 1 : Field Division into eight regions

Metrics Computed — preference and offensive value per progressive pass:

As stated in our methodology, we try to study the relation between preferred foot of a central defender and his/her passing characteristics. To do so, we use preference and understand the passing characteristics of a central defender, offensive value-added per progressive pass (OVPP) as our metrics. In this section, we define our metrics using some equations.
In the below equations, i denotes a particular attacking region of the pitch (as per Fig 1).

Fig 2 : Equation for Preference
Fig 3 : Equation for OVPP
Fig 4: Equation for value of an action
Fig 5: Equation for value of a game state

Visualisations and Inferences:

Now that we had our results ready, we plotted the marginal plots based on Preference and OVPP for all left centre backs. (Fig 6.1 — Fig 6.4)
The dotted lines in the figure denote the median values along preference and OVPP.

Fig 6.1 : Preference vs Offensive Value per Progressive Pass in Left Flank region
Fig 6.2 : Preference vs Offensive Value per Progressive Pass in Left Central region
Fig 6.3 : Preference vs Offensive Value per Progressive Pass in Right Central region
Fig 6.4 : Preference vs Offensive Value per Progressive Pass in Right Flank region
Fig 7: Median values for Preference and OVPP across different regions and preferred feet
  1. In the left flank and left central region:
    * Left-footed left centre backs prefer making more progressive passes than right-footed left centre backs.
    * Left-footed left centre backs also add more offensive value per progressive pass than their counterparts.
  2. In the right flank and right central region:
    * Right-footed left centre backs prefer making more progressive passes than the left-footed left centre backs.
    * Right-footed left centre backs also add more offensive value per progressive pass than their counterparts.
  1. On the right flank, there are few left-footed left centre backs that offer very high offensive value. These include:
    a. Santiago Selak (LCB — Genoa)
    b. Paul Dummet (LCB — Newcastle United)
    c. Aymeric Laporte (L_CB — Manchester City)
    d. Kevin Wimmer (L_CB — Stoke City)
  2. Similarly, on the left flank, a few right footed LCBs stand out with high offensive contribution. These include:
    a. Sokratis (L_CB — Borussia Dortmund)
    b. Unai Nunez (L_CB — Athletic Club)
    c. Waldemar Anton (LCB — Hannover 96)
    d. Michele Cremonesi (LCB — SPAL)
    e. Josip Elez (LCB and L_CB — Hannover 96)
    f. Javier Mascherano (L_CB — Barcelona)

Experiments:

We further try to utilise our analysis on some specific use cases —

Fig 8: Comparing Laporte’s (left-footed) and Otamendi’s (right-footed) preference and OVPP across different regions
  • Otamendi’s distribution of progressive passes is more uniform in Left Central, Right Central and Right Flank region while Laporte heavily prefers Left Flank and his preference drastically drops towards the Right Flank (roughly getting halved at each region) as shown in Fig 9.
Fig 9: Otamendi’s and Laporte’s Region Wise Preference
  • Also, Laporte provides significantly higher threat than Otamendi on the Right Flank (roughly four times) as per Fig 8.
Fig 10: Comparing Laporte’s (left-footed) and Ake (left-footed) preference and OVPP across different regions
  • Laporte and Ake highly prefer Left Flank to progress the ball and their preference drops drastically as they move to the Right Flank (roughly getting half at each region) as shown in Fig 11.
Fig 11: Ake’s and Laporte’s Region Wise Preference
  • Also, Ake, when playing as an LCB i.e. in a three/five defender formation, offers very similar offensive value distribution as compared to Laporte as shown in Fig 12. Both of these players generate their most threat on the Left Flank.
Fig 12: Ake’s (LCB) and Laporte’s (L_CB) Region Wise OVPP

Conclusion and Future Work

Based on our short analysis and inferences, we do find some strong patterns in ball progression and threat generation by central defenders based on their preferred foot.

Acknowledgement

We would like to thank Luca Pappalardo and Paolo Cintia for their valuable inputs and feedback for our analysis.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Shrey Grover

Shrey Grover

I love to read, love to write and in between the lines, I love to dream! I code too, sometimes.