Plotting individual playing time against goal difference

I started working with a new type of impact plot tonight, looking specifically at playing time compared against team goal difference. Dots representing each player are plotted along two axes: the horizontal axis records how much of the season the player has seen, while the vertical axis indicates the team’s goal difference during the player’s time on the field.

The following plot shows the Crew SC roster:

Columbus player performance, plotted by comparing playing time against goal difference
Columbus player performance, plotted by comparing playing time against goal difference

The dark gray dots show every player in the league, to help give a sense of context to each player. Steve Clark and Michael Parkhurst have played the entire season (100% along the horizontal axis), and unsurprisingly have a +4 goal difference – just like the team as a whole.

By contrast, Tony Tchani has played about 85% of the season, but the team has only managed a +2 goal difference in that span. Compare this with Mohamed Saeid, who has played less time (approximately 80% of the season) but against whom the team has earned a +5 goal difference.

Now, it can be dangerous to read too much into this type of analysis. With so few games played, there is a great deal of variability in opponents faced. A next step would be extend this to a teammate analysis, to see how given combinations of players correlate with team performance.

One last note about the Crew SC chart is that there appears to be two groups of players forming: the starters and substitutes. There is a group of 11 players that are clustering at the top right of the plot – the same eleven players who have started the last few games together.

With the team scheduled to host Seattle this weekend, I also looked at the Sounders roster. That plot is below:

Seattle player performance, plotted by comparing playing time against goal difference
Seattle player performance, plotted by comparing playing time against goal difference

Much like Columbus, this plot shows a grouping of preferred players at the upper right. One difference, however, is that the Seattle cluster shows a better goal difference, and is even more distinct from the bench players. Compare the distance between Osvaldo Alonso (60%, +6) and Leo Gonzalez (40%, +2) with that between Waylon Francis and Chris Klute on the Columbus plot.

One last plot before I wrap up for the night – this time, Real Salt Lake:

Salt Lake player performance, plotted by comparing playing time against goal difference
Salt Lake player performance, plotted by comparing playing time against goal difference

Granted, Salt Lake’s plot is distorted a bit by one outlier result: their 0-4 loss to New England recently. Even so, however, it is interesting to note that this team doesn’t have the “starters cluster” that is evident in the Columbus and Seattle plots. The data seems to indicate that Jeff Cassar hasn’t settled on a starting lineup that is consistently performing in the way that Sigi Schmid and Gregg Berhalter have.

It will be interesting to look at the Herfindahl Index for these teams to try and verify these assumptions about roster stability. This is something I hope to do in the coming days.

Leave a Reply

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