Racing Pass Networks with Arielle Dror

If you read this blog on a regular basis, you have surely read about my feeling about the teams collective passing performance and positioning being the key to their success.  I have recently become fascinated with the work of Arielle Dror on Twitter.  If you don’t follow Arielle, you absolutely should.  Arielle is s data scientist who works with American Soccer Analysis.  If you are a regular Twitter user you are probably familiar their data being used for g+ GameFlow charts or many other xG models.  Since Benton from VamosMorados.com has covered xG in depth, I wanted to focus on the charts that Arielle creates. 

I have mentioned on this blog at least a few times and definitely on Twitter that I am not a big fan of xG.  I am not a huge detractor of the measure either, I just don’t find the way in which it is typically presented particularly useful if you watched the match. (Almost everyone else I respect and have spoken to about xG like Arielle, Benton and Zach-Allen Kelly disagrees). The exception to this is Arielle’s xG “Race Plots” which are presented on a graph with match time on the x-axis and cumulative xG on the y-axis.  Her graphs are annotated with the actual goals shown at the point in time with the corresponding xG value for the shot that led to the goal.  The other cool feature is the win probability chart on the bottom.  I believe that Arielle told me that this was based on a set number of simulations based on the xG values from the match.  Here is the xG plot from Racing’s most recent match:

You can see that Racing had a 5% chance of winning the match based on simulations using the xG values for the shots in the match. All of this is a bit of a preamble and context setting to get to what I really want to talk about and that is Arielle’s fascinating Pass Networks.

If you have ever worked in a technology or technical/process discipline you have probably come across a network diagram.  These diagrams are often shown to demonstrate “flow”.  That flow can be materials, people, knowledge, data, etc.  Several analysts and data scientists like Arielle have started to use network diagrams as a template for Pass Network visuals. These charts are a fascinating way at examining a team’s positioning and passing.  Let’s take a look at the Pass Networks from the same match.

The Spirit’s Pass Network is typical of a team that performed well in my opinion.  Pass Network are built using goals added (g+). These values measure a player’s on-ball contribution in attack and defense. For more on g+ visit American Soccer Analysis here. The nodes on the charts represent the players and are based on their “average” position on the field when getting a touch.  The size and color of the node are based on the player’s frequency of touch and the value of those touches. “Edges” are the line segments and represent connections of greater than 5 passes, with the thickness of the edge representing the frequency. Now I want to venture into the unscientific part of the analysis.  Most people will tell you there is a bit of art and science in any analysis.  To my mind, a Pass Network should mimic your formation and playing philosophy.  For example most Pass Network charts that I have seen have a nice diamond or sideways “kite-like” shape.  I wouldn’t go so far as to call this optimal, but it is definitely typical. Examining this particular Pass Network tells me a few things:

1.      The Spirit preferred their attacking right hand side.  There are more nodes there and the widest nodes are on the bottom half of the chart.

2.      Several players completed over 5 passes to a wide variety of players.  For example, Kelley O’Hara had a connection with 7 other players where she passed or was passed to by the other player.

3.      I may have under valued Trinity Rodman’s contribution to the Spirit.  While I still think she need to work on her finishing, she was a vital cog in the build up to the goals.  The red color on her node is a positive value and is due to her two assists in the match.

As a point of contrast here is Racing’s Pass Network from the same match:

Here are the things that jump out at me from this chart:

1.      Erin Simon and Cheyna Matthews basically only received or passed the ball through single routes with Kaliegh Riehl and Yuki Nagasato respectfully.  Any connections with other players would have involved fewer than four passes.

2.      There are much fewer passing triangles in the Racing’s chart.  Almost every Spirit player was involved in multiple passing triangles.  In Racing’s chart none of the forwards was really involved in a passing triangle at all.

3.     Racing’s chart looks disconnected at points, demonstrated by a couple of edges that dead end.   Washington’s chart seems connected at all points.

4.      There was virtually no passing in the direction of play that I would classify as wide to center. Effective play in my opinion utilizes playing the ball from wide players into the central ones especially in attack.  Look at the chart below from Racing’s match vs. North Carolina.  You will see what effective wide play looks like, especially from Carson Pickett.  You will also notice that even though she is at left back, Carson’s average position was in the attacking half.

For a comparison, here is Racing’s Pass Network for the same game and the less said about it the better:

These are basically pictorial representations of a few of my main issues with Racing’s play.  They don’t utilize the width of the field effectively, especially in attack and their forwards are often isolated and not involved in much build up play.  In Racing’s better performances, you can definitely see better Pass Networks.  Let’s take a look at three of Racing’s better performances:  a 0-3 win at Chicago, a 1-1 draw in Orlando and a 1-1 draw at home vs. OL Reign.

Let’s start off with the win in Chicago:

Honestly, this is a peculiar looking Pass Network at first glance.  It looks like all of the play from the back came via the Martin/Fox/Olofsson connection.  It’s definitely not a typical look of a winning Pass Network performance.  On the other hand, there was definitely a direct Bonner/McCaskill/Millet connection that seemed to be effective in cutting through the heart of midfield and ignoring Racing’s weakness of wide play in the attacking half. Savanah McCaskill had the most valuable touches in the match which is denoted by the red-orange color of the node (1 goal, 2 chances created). Arielle pointed out to me the number of edges (7) coming from her node. This is a good example of a player dictating play.

For the 1-1 draw at Orlando:

This is a typical Pass Network that you seem from most teams.  There are quite a few triangles and virtually no islands, bar the late appearance of Julia Ashley.  You might also notice that a deep dropping Salmon was more involved that when she plays further up the field.

The last Pass Network is from the 1-1 Draw vs. OL Reign on September 4th.

Again, you will notice Salmon on her lonely island.  Jorian Baucom, while connected seems only connected via two players.  This Pass Network probably best shows Nadia Nadim’s impact.  Her average position is in the 10 spot on the field.  However, there are still quite a few disconnects in the network.

So, what is the take away from all of this?  For me it would be interesting to know if data like this is being used by the team to evaluate performances.  I think it is a great way to quickly see where your team’s passing strengths and weaknesses lie.   For Racing, I hope that at least if the team is using this particular view, the analysis of passing will lead to an improvement in play next year.  

I would like to say a very special thank you to Arielle Dror who supplied all of the charts in this post and who was kinds enough to review my thoughts and suggest a few edits. You should definitely follow Arielle on Twitter @arielle_dror.