Explainer: How to read my graphics

How to look at radars and crab cakes

I get the question from time to time about the graphics, “How do I read and make sense of this” and it is a fair question to ask and one that I will look to bring up here.

The TLDR: answer is if the graphics are more filled out that means that player ranks highly compared to peers in many statistical categories.

For a more detailed explanation let’s continue below:

So I will typically use a few kinds of graphics for a player the first one and the one that has been long associated with StatsBomb is the radar and my version looks like this:

The top section gives information about the player: team, league, minutes, age, position (this is who they will be compared against as well), and the season.

The middle section is the visual representation of the player’s performance. The stats represented here are based on the position (and I have debated what to include heavily) and I think try to give a balanced view of the main things that the position is involved in and a team might be interested in caring about.

The population for the comparison is players that played at least 60% of their minutes in the position while also playing at least 10 90s. I use players from the “Big Five Leagues” and have data going back to 2015-16 for these, and Premier League going back to 2011-12. These players are used to set the percentile rank for the stats used and the scale on the chart. The scale on the chart is the 95th percentile for the outermost ring and the 5th percentile for the innermost ring with a linear scale between these two points (not the middle of the radar would not always translate to median because the distribution on these is not always like that). This helps so that some of the true outlier-type players don’t massively skew how things look and help to be able to tell by looking where generally a player ranks.

UPDATE: For stats where having less of them is better (turnovers, dribbled past, fouls committed, etc) these are presented reversed. For these the higher value will be on the inside ring and the lower value will be on the outside. This is done to help ensure that at a glance further from the center is still considered “good”.

The bottom part gives the per 90 values for the stats and the percentile rank against the same population.

The next graphic was popularized by Football Slices (RIP) and is a different spin on the radar using percentiles instead of a linear scale of the radar.

For this, the information at the top, and the filtering for the population are the same as on the radar.

What changes here is that instead of having the data presented with a scale based on the 95th percentile to 5th percentile and a linear scale in between what happens here is that it is simply just the percentile rank for the player and it runs from 0 to 100.

The bar that is presented is percentile rank, with further from the center a better rank to be able to keep the more filled out is a “better” look. I have added in the per 90 stat in a box along the outer edge as well to help display the actual production as well.

UPDATE: Just like for radars, for stats where having less of them is better (turnovers, dribbled past, fouls committed, etc) these are presented reversed where being lower will be the top of the percentile rating instead of the bottom. This is done to help ensure that at a glance further from the center is still considered “good”.

On the slice chart, it is a little easier to add more information so these will tend to have a bit more on them compared to radars. To help make interpreting a little easier I break them into three general categories of attack, possession, and defense based on color as well.

Interpreting Radars and Crab Cakes/Slices

The main case for radar/slice charts is that they are kind of easy. More filled out generally corresponds to a better player.

As you look at more of them and get a feel for where the stats are you can at a glance get a sense for what sort of style the player has and you can do it pretty quickly because humans are really good at noticing patterns visually.

Not all metrics are quality based, with some measuring style and how they are used tactically by the teams. For midfielders and defenders, I use a possession adjustment, denoted on the charts as Padj. (this is based on the speed each team attacks at and the number of opportunities that would entail for making defensive actions).

The stats I use are all normalized to a per 90 basis or a percentage of the relevant stat that is being done. This is done to account for different playing time between players, I also have a fairly high cut off for players to be included and if a player’s radar is created that falls below the minutes a “Small Sample Size” warning is displayed.

I have five templates currently that I have as the default (anything can be customized as well and I am playing around with thinking about more role-specific stuff but that’s still a work in progress).

I have Forward, Attacking Midfield and Winger, Midfield, Fullback, and Center Back.

For the different positions, I feel most confident that for attacking players the stats give a very good sense of good stats = good player. Attacking metrics are generally well-calibrated with the goals of what attackers are trying to do.

For midfielders and fullbacks it might be a little less clear because these are often even more dependent on what the team is asking these players to do and with the stats still generally calibrated to capture the intent of what a player is trying to do in the role but it can be more of a step back and using a sort of proxy to gauge how well it is done.

Center backs are the hardest and the one that should be taken with the biggest grain of salt. The possession type stats I think work fairly well but the defensive stats describe what a player did and how the team plays but I think doesn’t do a great job of describing the quality of the player.

Defensive event stats is absolutely not something that I would put a lot of stock in telling me if a player is good or bad but rather telling me what actually happened on the pitch but even then it is something that needs to be dug into more. The big issue is that with defense we are trying to measure the absence of something occurring (mainly goals, but also good quality shots, or dangerous possession) and that is not always compatible with event data that only has something to record when an event happens.

If a player is in a perfect position so that an opponent doesn't try a pass that could lead to a goal, that is something that isn't measured and no one gets any credit for. On the flip side a player that screws up his positioning but makes a good recovery tackle gets credit even if the only reason he made that action is he messed up at first. As you can see it can get messy really fast.

—

Let me know if you have any other questions or things that should be updated in the comments below. (these will be open to all)