Clustering Goalkeeper Distribution

In the previous post I used StatBomb’s play by play World Cup database to calculate expected saves. After looking at the database more, I noticed that every pass is also logged, meaning a goalkeepers distribution performance can be measured as well.

Based on every passes pass length, pressure, height of pass, and part of body, I am able to cluster every pass made by every player as easy, medium, or hard. I am left with a table that looks like this

Every pass is documented by goalkeeper name, match id, completion, and difficulty level

Based on number of passes attempted, and completion percentage within each pass difficulty level, I am able to give each player a rating on their distribution performance. This rating also takes into account number of games played, and finds average rating per game.

A list of all goalkeepers, in alphabetical order, with their rating/game at the 2022 FIFA Men’s World Cup

After ranking the goalkeepers on rating Manuel Neuer and Devis Rogers Epassy Mboka had the top two ratings, while Allison Becker and Steve Mandanda had the bottom two. We can visualize all four of their distribution performances below.

We can see that the model doesn’t punish a high amount of difficult passes attempted, as long as the keeper is converting 40-50% of them.

Conversely, the model harshly punishes easy and medium passes that are not completed.

By combining this statistic with our expected saves statistic from earlier, we are able to get a wholistic understanding of every goalkeepers performance at the 2022 FIFA Men’s World Cup.

Rating represents their distribution performance rating and Difference represents their made saves minus expected saves.

We are able to easily see the elite performers at this World Cup, by looking in the top right, and the poor performers by looking in the bottom left.