Player profiles: passing

Introduction

A useful outcome when analysing sports is being able to categorise, or group similar, players. I hope to make this post the first part of a (possibly non-consecutive) series where I endeavour to do this, eventually trying to use cluster analysis to be allow players' style to be effectively compared. The first area of play considered is passing. For each area of play, the following factors will be considered:

• Involvement - Relative to the team average, how frequently is the given player involved in this area of play
• Location - Where on the pitch is this player most often involved
• Behaviour - The least clear cut. Put simply, this is what the player attempts to do in this area of play

Note that success/failure in the given area of play is not considered (for passing, I have previously discussed a more useful metric for passing success). The aim of this project is merely to group by style of play: it is possible for two players to have similar style but one to be much better than another. The converse is also true: two players may have contrasting styles but be equally effective.

 Involvement

For passing, the involvement factor for a given player is calculated by:

Involvement = Player passes attempted per minute / Team passes attempted per minute

The "per minute" factor for the team is total minutes (i.e. for each player, so in a match where no player is sent off, 990 total minutes are played by each team).

This factor is scaled by the team value to ensure that players who play for teams who typically have less possession are not penalised for that - their involvement should be judged relative to that of their team-mates. Likewise, players for teams with high possession percentages should not have their involvement factor inflated by that.

The players who fall in the top 20% by this metric are described as "focal"; those in the bottom 20% are "peripheral".

Location

For passing, the location factor for a given player is calculated by:

Location = (Final third passes - Defensive third passes)/Total passes

 In this case, the more positive the value, the higher the proportion of passes played in the final third; the more negative the value, the higher the proportion of passes played in the defensive third.

Again, the top and bottom 20% are given descriptions, which in this case are "attacking" and "defensive", respectively.

Behaviour

There is no single measure of behaviour - it is split as follows:

• Distance of pass. John DeWitt and Paul Quinnell have previously looked at long passing
• Direction of pass
• Forward/backward
• Left/right

The distance of the pass is given by:

Distance = Number of long passes/Total number of passes

The larger the value, the higher the proportion of long passes played. The smaller the value, the lower the proportion of long passes played.

The top and bottom 20% are described this time as "long" and "short", respectively.

The forward/backward metric is calculated by:

Forward/Backward = (Forward passes - Backward passes)/Total passes

A larger value indicates a higher proportion of forward passes.

The top and bottom 20% are called "forward" and "backward", respectively.

The right/left metric is calculated by:

Right/left = (Right passes - Left passes)/Total passes

A larger value indicates a higher proportion of passes to the player's right.

Using the categories

A glance over the results confirms that these categorisations seems to agree with expectations. Observations include:

• Regular first team players (more than 1800 minutes) who fall into the "focal" category are: Charlie Adam, Joe Allen, Mikel Arteta, Benoit Assou-Ekotto, Gareth Barry, Joey Barton, Chris Brunt, Yohan Cabaye, Michael Carrick, Cheik Tioté, Peter Crouch, Alejandro Faurlin, Marouane Fellaini, David Fox, Karl Henry, Karl Henry, Bradley Johnson, Frank Lampard, Raul Meireles, Luka Modric, James Morrison, Youssuf Mulumbu, Danny Murphy, Samir Nasri, Steven N'Zonzi, Morten Gamst Pedersen, Aaron Ramsey, Angel Rangel, Nigel Reo-Coker, David Silva, Alexandre Song, Adel Taarabt, Rafael van der Vaart, Glenn Whelan, Dean Whitehead, Ashley Williams, Yaya Touré.
  Mostly, these fit in with what might be expected from watching Premier League football last season
• Of the 23 regular first team players who are classed as "long" and "forward", 3 each play for Aston Villa, Bolton Wanderers, Everton, Stoke City and Wigan; 2 each for Blackburn Rovers, Norwich City, Tottenham Hotspur; 1 each for Sunderland and West Brom
• Approximately 11% of players have no outstanding passing characteristics (in the middle 60% for each categorisation)

The main value in this categorisation is in discovering players whose profiles are the same. In some cases, this is unsurprising, but in other this is not true. Consider, for example, David Silva. He can be described as focal, final third, short. There are only 2 other players who may be described as such. One is Samir Nasri, which is not necessarily surprising. That the other is Peter Crouch is more of a surprise, since these players have contrasting styles. This may indicate that further segregation is required, for example taking into account headed passes.

A further example of non-obvious grouping of players is seen when Stephane Sessegnon's category is considered. He is classed as final third, backward, left. There are only 2 other regular players who fit into this classification, but both are also classed as "peripheral". These players are Mario Balotelli and Gabriel Agbonlahor.

Conclusion

I do not believe I have yet fully realised the potential of this kind of categorisation of players. Simply listing interesting examples does not do justice to the categories which have been created. I have already discussed how this work could perhaps be advanced by including further passing categories, for example flick-ons and/or crosses.

In my opinion, with appropriate modifications, the process described in this post will provide useful information about the passing profile of a player. In addition, when this work is furthered by profiles of other aspects of play, for example shooting, defending and maybe set pieces, the overall profiles generated will accurately define the style of play that each outfield player in the league favours: an outcome which will be of great use to viewers and teams alike.

Please get in touch with comments or suggestions, either below or on Twitter (@hpstats)