Introducing Marking Maps

Intro

One of the most exciting aspects of working with tracking data is the possibilities it opens for analyzing defensive performance. While you can get quite creative just by looking at events, some things won’t be captured or will require a huge sample.

With full match data, you can now look at defensive positioning and marking in a single march. So, I guess it’s time to make my own defensive metric…

In this post, I’ll look at how we can infer and visualize Marking relationships by looking at tracking data.

A sneak peek, just to make you keep reading — because no one likes walls of text

The Data

With the release of Metrica’s tracking data samples, we can now experiment with the high quality full match data necessary for this analysis. For this post, I used Period 1 of Metrica’s Sample Match 1.

From the data I created two subsets: for each team I gathered all the sequential events from that team and took the first and last frame from the possession chain.

Possession is messy in football, so a better definition would be helpful. But for now, I’m trying to keep it simple.

The Definition

The marking relationships were calculated as follows:

1. Calculate the ideal position for marking each attacking player (defined as two meters from the attacker towards the goal)
2. Take a defender, and then:

• Calculate the distance from the defender to each attacker’s marking position.
• Filter to the minimum distance.
• If that distance is above a threshold (I used 7.5m), then assume the defender isn’t marking anyone.

3. Loop through all defenders (so, one defender can only be marking one attacker, but two defenders can be marking the same attacker).

4. If an attacker has no defender marking him, add a row to the table (with no marker).

There are a lot of things you can do to improve on this (e.g. consider momentum, prioritize free attackers, prioritize stable relationships). That may be a subject for a later post.

The Viz

The images below show a few random moments from the match. The marking relationships are shown as a line between the defender (with a circle around him) and the attacker:

The results seem reasonable. There are a couple of instances where I could have added or changed a relationship (e.g. in the second frame, is #19 marking #5?), but overall it looks fine.

Note: while these are pretty much random, I specifically avoided frames near set pieces, because they’re a lot more confusing. Something to be addressed.

The Map

Inspired by the more common Passing Map (and suggested by @johnspacemuller), I tried to make a Marking Map.

A typical Passing Map

I plotted the average position of each player and added a line for each common marking relationship (I defined common as being present at least 10% of total possession time). The thickness of the line represents frequency.

Red Team Attacking

Marking Map for red team’s possession

What can we take away from the map:

• Overall, the teams look balanced.
• The fullback seem free to help the center defense, while the wingers cover the opposing wingbacks.
• As an attacking strategy I would look into overwhelming the central midfield with a forward run from either the DMC or one of the CBs.

Blue Team Attacking

Marking Map for blue team’s possession

Now this is much more interesting:

• the centerbacks and the defensive midfielder mark the two forwards (one of which drops deeper)
• the wingbacks cover the opposing wingers, but #18 is on average higher up the pitch than the winger
• they’re helped by a central midfielder (on the right) and a centerback (on the left)
• note how the blue team has interesting variations between the defensive shape and the offensive shape: #7 goes left, #10 drops deeper.

This is the from the buildup to one of Blue team’s shot, showing what we saw in the map: a 2 vs 2 counterattack resulting from #18 not defending and #10 joining the winger (#8) on the right.

This isn’t a robust result yet, but it’s a promising direction. Let me know what you think over Twitter.

Future Work

This is the first post in (hopefully) a series about analyzing defensive performance using tracking data.

Topics for future posts:

1. Better definition of Marking
2. Deeper analysis of Marking relationships
3. Creating metrics for quality of Marking
4. Analysis of Marking during Set Pieces

This work could only be done with the quality full match tracking data provided by Metrica Sports. They are currently looking for partners who can provide the rights for the release of a more complete set of data. Please support this initiative if you can.

Project: Football Open Data -