There has been a lot of work recently looking at teams and players performances based over a full season or several seasons however how do we measure a players in season performance to determine if they have actually improved game on game or just had a one off? This is probably one of the key aspects for a manager in order to determine who to play, sell and buy. It can also act as an indicator as to whether specific training programmes are working or if there are specific weaknesses that need to be worked on. Finally by being able to quantify performance it allows the manager and coach to be more objective in their decisions.
To develop a method of quantifying a players season on season performance level.
A method which is applied in sports science to assess an athletes performance level is to measure their mean (average) performance (time, passes complete) over a number of tests (races, matches etc…) and then work out what an individuals Typical Error (TE) is in their performance. By finding the mean we can now quantify what an athletes average performance should look like over a number of races or matches. Once we have the mean and have started to plot the athletes performance over a season it is more than likely you will see some performance above and below the mean. Some performances will massively fluctuate from the mean while the majority will probably hover around it.
To determine if a performance was a fluke or something to get either very excited or deeply worried about we need to understand the athletes TE of performance? We calculate the TE by dividing the Standard Deviation (SD) of the performances (the deviation of scores from the mean) by Square Root (SR) of 2.
TE = SD/(SR2)
Now we know the TE we can determine if a performance which deviates from the mean is just an expected variance in performance or something that is above what we would expect to be the norm. For more detail on this look at Will Hpokins website SportSci.org or check out this presentation here.
For our example we are going to look at the passing completion rate (PC) of Sebastian Larsson to determine if he has had an improvement in the number of passes completed between the 2011/12 and 2012/13 season. To do this I collected Larssons completed passing stats via StatsZone for last season and this.
To determine if there had been an improvement in Larsson’s pass completion rates from the corresponding games I calculated the difference in performance (i.e. PC vs Liverpool Away 2011/12 – PC Liverpool Away 2012/13). Once we have the differences for each game we can now plot them into a Bland Altman Plot to visualize the difference in performance.
While this provides with a good game on game overview of Larssons’s passing performance it doesn’t allow the coach to see whether his performance has been of a greater benefit to the team. To help this we can then look at individual matches and use a spider-graph to show the players performance in specific key performance indicators.
This now provides the coach with both a season long view and in game view to make a more objective assessment of Larsson’s performance improvement.
Results – Assessing a Players Passing Performance
The above graph has put into practice what we discussed above looking at the season on season passing performance of Sebastian Larsson. Straight away we get a visual indication as to whether Larsson has performed at a higher or lower level to the same match of the previous season.
The red lines show the TE we would expect Larsson’s performances to fluctuate between which is around 12 passes +/- anything outside the TE would indicate either an exceptionally good or poor performance. We can also judge whether that performance is a one off or whether the player is consistently out performing their previous season. Looking at the graph we can see that Larsson is consistently out performing his previous seasons passing stats except for the game against Liverpool which coincided with a 3-0 loss and a slight under performance against Aston Villa which also result in 1-0 defeat, however both of these performances fall within the expected TE.
What is impressive is the number of performances which have seen Larsson out perform his previous season’s passing stats that are over the line of TE indicating on initial inspection that his performance levels have improved. Looking in slight more detail this may be down to Larsson now being deployed by Martin O’Neil as a central midfielder rather than winger which you would expect to see more of the ball.
By breaking down Larsson’s performance into a spider-graph we can now determine if his improved passing performance against Newcastle for example has been of a greater significance. As can be seen Larsson’s forward pass completion (PC) has significantly increased indicating that he has been more influential in creating attacks which is further supported by an increase in the number of chances created and crosses provided. As a manager O’Neill could deduce that moving Larsson to centre midfield has had both a positive impact on Larsson’s influence on the game and also the performance of the team.
The aim of this post was to introduce a different way to assess a players performance levels using TE and Bland Altman plot which are common place in sports science. The graph backed up with a more detailed spider-graph seems to be able to provide good data which would be useable by a coach and manager to objectively determine the level of performance of their player. In this case we looked at Sebastian Larsson who has demonstrated a consistent improvement on his previous seasons performance indicating that he is having a better season.
I plan to revisit this type of analysis more often.