Attacking Styles and Defensive Weaknesses

An Expected Goals Model works by categorizing chances and assigning a value to each category. In most cases we just add up all Expected Goals for a team or a player to measure performance, but it’s also interesting to take a closer look at those categories themselves.

I’ve looked at a few recognizable ‘types’ of chances (based on shot type, location and assist type) and asked myself the question if a team’s attacking style can be identified by the types of chances they create, and similarly if teams have certain defensive weaknesses against specific types of chances.

In other words: how much do chances of a certain type contribute to the total ExpG or total conceded ExpG?

Here are the numbers for the Premier League as a whole over the last four full seasons:

totals

To see if these numbers are actually meaningful I’ve looked how much they differ per team (relative standard deviation), and to what extend they are repeatable (correlation between the first and second half of a season). Any stat that tells something about a teams style should be repeatable at least.

The first thing you’ll notice is that repeatability can hardly be found in types of chances conceded. Teams play to their own strengths much more than they play to their opponents weaknesses. This actually surprised me a bit. I would suspect that if it’s known that a team has trouble defending crosses other teams will use that knowledge, but it might be easier said than done if you don’t have the players for it. On the other hand, a manager can change a team’s attacking style as we will see later.

A couple of examples:

setpieces

Teams that are consistently weak at defending set pieces? There’s no such thing aside from one exception (the top right outlier): Arsenal during the 2010/2011 season.

Penalties are all over the place and completely random:

penalties

Here you’ll see that the creation of a certain type of chance is more spread out and shows more correlation than the amount of chances conceded of the same type:

headers

throughballs

All in all I would say that there are only three numbers that are definitely meaningful: chances created from through balls, headers and shots from outside the box. Between through balls and headers there’s also a negative correlation of 0.54. Without a doubt we’re looking at different attacking styles.

Here’s a view of this season’s data which I like to call arsenewenger.png:

arsenewenger.png

A closer look at Arsenal shows that although he fits right in, it’s not just Özil either. Here are the top ten seasons (out of the last four) in terms of %ExpG from Through Balls:

toptb

You want more manager fingerprints? Here’s the full picture from this season:

allteams

Notice Crystal Palace as the leading team when it comes to headers? It’s no coincidence. Over the last four seasons under Tony Pulis, Stoke averaged more than 30%. Now Pulis gets the Palace job and immediately their percentage is up to 32.7% from 24.6% under Holloway.

And then there’s Swansea, currently the team with the lowest share of through balls. If they ever were a poor man’s Arsenal they’re not doing a very good job now. Under Brendan Rodgers in 2011/2012 they managed 11.1%, in the first half season under Laudrup it was even up to 11.8% but then it dropped to 4.6% and all the way down to 0.7% now (that is one shot from a through ball all season). This blows my mind as Michael Laudrup was an absolute master of the through ball himself. At the same time you can find Rodgers’ first season at Liverpool in the top ten above, right there as the highest non-Arsenal team.

That leaves us with one high profile managerial change which doesn’t show such a clear picture. After David Moyes’ move from Everton to Manchester United, Everton’s headers are down slightly and shots from outside the box are up (Barkley, Mirallas), but it’s not a huge difference. United’s through balls are down a bit, and headers are up, but that trend was already going on under Ferguson:

manu

 

Leave a Reply

Your email address will not be published. Required fields are marked *

+ 26 = 35