I started producing these radars – god, on Monday? This week feels like a year. As mentioned in the initial post, they are something I’ve wanted to do for months, but I didn’t have the data collected yet or the skill to do them. The skill element of these is still in question, but I have gathered a fairly comfortable data set from 09-10 from the many public sources out there.

One of the things that occurred to me while building the base template was that these would likely be much more useful for specific positions. We don’t expect defenders to take 8 shots a game like Ronaldo, and it’s pretty silly to expect Ronaldo to make 5 tackles a game like Arturo Vidal. So that’s what I started doing after publishing the Ballon D’or piece.

The first position template I built was for central midfielders. I wanted something that reflected the fact that CMs are potentially the most versatile players on the pitch. They need to attack and defend. Some play a big role in the offense, contributing high levels of key passes, throughballs, some goals, dribbling, etc. Others are destroyers that sit much deeper, and recycle the ball quickly back to offensive players. I figured I could create a radar that reflected the full potential of central midfielders.

This is what I came up with. [For those wondering about design decisions, I’ll discuss it a bit down at the bottom. I am aware there are imperfections in displaying information like this.]

Mikel_Arteta_201213

That’s Mikel Arteta from last season. His role is very defensive, with very little area covered on the offensive side of the ball.

Now for someone a little different.

Cesc_Fabregas_0910

That is Cesc Fabregas at Arsenal in 2009-10. I have stated for quite a while that I think Fabregas is the best attacking passer I have ever seen in the Premier League. Given his key pass and scoring contribution (NPG+A90), that seems pretty accurate.

One of the cool things I was hoping to do was use these radars to display how role changes alter statistical outputs. Check this out.

Fabregas_Arteta_Transparency

Same chart, two outstanding players, two completely different shapes. Cool, right?

Now how about this? Here’s Aaron Ramsey for Arsenal so far this year (For those reading from the future, the data was good through Jan 17th).

Aaron_Ramsey_2014part

Ramsey probably plays a role much more comparable to what Fabregas played when he was at Arsenal than Arteta, but obviously they aren’t the “same.” They have similar defensive I+T but Cesc intercepted the ball more and Ramsey tackles more. They even have similar scoring contributions (NPG+A90), but Cesc’s were split pretty evenly across assists and goals, while Rambo has scored a ton this year. Fabregas is the more creative passer by far. Now let’s see what happens when we overlay these two.

Ramsey_2014_Cesc10_overlay

 

I’m going to keep developing the positional radars based on KPIs I’ve examined and think are important, some of which will obviously get adjusted after feedback. Some of the contextual ones like Dispossessed, Fouls, Was Dribbled make people shrug, but they are actually fairly important bits of information when evaluating midfielders. Other positions will have different axes, and presumably different contextual stats to go along with them.

If you like these – and response from most people has been very positive – then check back more in the coming weeks for new positions and comparisons. I’m still very much learning with these, but one thing I am hoping to look at is what happens when you overlay entire midfields on top of each other and then compare them across teams.

Bonus Radars

Michael_Carrick_201213

Vidal_DM

 

Design Decision Discussion
Though I am new to creating these, I have been studying this for a while and am aware of many of the pitfalls for both the data itself, and then method of display.

  • The boundary choices for each axis are the actual boundaries that a player has produced in that position for the data set. So in today’s case, a player has actually produced those numbers playing as a midfielder across a reasonable chunk of one season.
  • I have grouped similar attributes together. Passing accuracy is neutral, but tied to the passing stats. Long balls are attached to it, but lie near the defensive metrics as well (typically defenders play many more long passes than offensive-focused playrs). Scoring and offensive metrics are to the right of the chart, while contextual metrics are down at the bottom. Defense is to the left. Thinking about it, Dispossessed should likely be next to Successful Dribbles. I will change that in a future version.
  • Good performance should take up more area, so the “bad” contextual stats have had their scales reversed. If a player was never dispossessed, they would hit the outside of the chart. Never dribbled past? The same. Basically, players with the best stats should cover the greatest area in the radar.
  • Sobhan

    I guess we need two separate radars for positive and negative stats for a particular position. The circle can possibly be divided into two separate semi-circles with positive & KPIs on opposite sides.

    Great post, nice idea.

    Thanks.

  • bozz

    Great read, Ted. Nice job!

  • toshack

    This is cool Ted!

    If you build a radar image of other postitions (which I’m sure you will) and keep the same amount of KPI’s you could actually do a overlay of mulitple players within a team (maybe even the entire team) to get a picture of the “team coverage” (or only defence players or only offensive players). The bigger the coverage as a whole, the stronger the team (at least in theory and statistically). Just a thought.

  • http://twitter.com/Von_Sparron von_sparron

    Very cool! This would be a good tool to see how a player’s game changes over multiple seasons when playing in different positions. For example Cesc at Arsenal compared to Cesc in the early seasons at Barca compared to this year at Barca.

  • MJ

    i agree concerning the placement of “was dribbled” i think if you switched this with “dispossessed” and had it go Successful Dribbles, Was Dribbled, Fouls, Dispossessed, Tackles, etc.. i think that would make more sense…thanks for this! excited for more!

    • MJ

      sorry meant Successful Dribbles, Dispossessed, Fouls, Was Dribbled, Tackles…keeping was dribbled in between fouls and tackles because, when someone is dribbling at a defender, those are the 3 possible outcomes – sorry for confusion in previous post – cheers!

  • Leroy Cain

    I like these radars, Ted. Good job.
    Arteta & Carrick pretty darn similar! Will be interesting to see who covers the most area in these and comes out as the best all rounder…?

  • John Shiels

    Brilliant work, gents. I dread to think the work that went into them but they are going to be very useful and they look sooo pretty, too!
    Just wondering if you’ll be using averaged stats for the entire midfield/defence radars or if you have some other cunning plan. I’m sure I’ll find out soon enough…

  • Morten Dalbøge

    Great work, I love the way you break down the game.
    A couple of thingd I find is missing from the chart:
    1) Distance covered
    2) Amount of forward passes
    3) ball possion or touches

    With out these, it makes no sense comparing the pass completion rate. Likewise the statistical value of a player that looses possesion of a ball makes little sense if he only has it a few times during a game. this is usefull when comparing Rambo and Wilshare.

    Keep up the good work

    • MattV

      Losing possession can be very influential, even if it happens only rarely. Ramsey memorably lost possession just outside his own penalty box in one of the group games against Dortmund, from which Mkhitaryan (sp?) immediately scored.

  • craig

    This is superb stuff.

    Re: choice of boundary scales, you may find it useful to calculate them as z-scores.

    Z score = individual score – mean score / standard deviation of mean score

    Positive values would indicate players score is above the mean and negative values below the mean.

    Worth a try to see if it adds value.

    • tknutso

      Tried it. Really dislike the loss of informational display. Basically, you take it out of the layman realm and into the geeks-only realm at that point.

      • toshack

        Good call Ted (with all due respect to Craig),

        What is sooo good about the radar charts is that they are intuitively understandable however much data and stats are behind them!
        You can discuss what KPI’s to include or not but this form of presentation is really on the forefront of things (of course the internal placing of the KPI’s matter as you have pointed out).

        (Besides I read the basics of Statistics at University a looong time ago and have worked hard to forget it ever since, so appreciate the layman approach…)

  • Steve

    Radar Love.

    One suggestion, could the background be divided into shaded areas that correspond to Offense, Passing and Defense? In that way you can see at a glance the tendencies or focus of a player especially a midfielder.

    Steve

  • ThreeFourThree

    Love it. I honestly wouldn’t change a thing. I can’t wait to see the radars for wingers and attackers. Given the amount of teamwork that goes into defending, I am suspect of how good these will be for rating defenders. I’ll still read the article though!

  • Ian Baldwin

    Hi Ted,

    I was just looking at the KPIs you were using and was wondering whether there was a more useful metric available by combining ‘Tackles’ (T) and ‘Was Dribbled’ (WD)?

    I was thinking of something along the lines of either ‘Tackle Score’ (TS) or ‘Tackle %’ (T%), where:

    TS = T – WD
    T% = T / ( T + WD )

    or this this end up oversimplifying tackling by viewing every tackle as stopping an attempted dribble and ignoring a players ability to close down opposition and catch them dawdling in possession?