Wednesday, 19 September 2012

Quantifying Passing Difficulty in the EPL.

Goals and goal attempts are quite rightly the standout events that occur during a football match and their relative scarcity allows for extensive analysis to be undertaken on the easily collected data. In contrast passes are much more numerous in game events and while this gives us a much larger dataset to work with, the collection process quickly becomes much more problematic. Therefore the recently released xml data from last season's Bolton verses Manchester City Premiership match has provided an ideal opportunity to work with a substantial amount of quality passing statistics.

Analysis of passes attempted has quickly evolved from the bland overall completion figures, to area and directional subsets and is now moving towards investigations based on each individual pass. Unlike goal attempts, where the point of origin of the attempt is usually sufficient to make analysis possible, passing data also requires an intended end point. We can then incorporate these parameters into our regressions and begin to estimate an expected pass completion figure based on field position and direction and difficulty of the attempt.

Ideally this type of analysis needs to accumulated passing data from every team in the league in order to provide an average baseline for the expectancy figures. However at the moment we are restricted to using albeit extensive data from a single game. City and Bolton do have contrasting passing styles, the former ended the season as Champions, the latter were relegated, so there is a possibility that the pass expectation figures derived from the data of both teams may approximate to Premiership league average values. However, we are putting convenience above rigour at the present.

If we firstly perform a general regression that generates the likelihood that a pass will be completed using the co ordinates of the starting and end point as the four inputs, we can try to see which of the two teams was better at pass completion when account was taken of passing difficulty.

Predictably, Manchester City attempted more passes than Bolton, but they also completed more than our hoped for "average team " would complete. 427 found their target compared to a cumulative expectation of just 410 completed efforts. Inevitably, Bolton therefore appear to under perform completing 279 passes instead of an expected 295. Despite the problems associated with our methodology these conclusions aren't unexpected, as City were top of the pile in May and Wanderers were relegated on the last day of the season.

To see if City's superiority is maintained in different areas of the pitch, I've then looked at their completion rates for passes made from the final third of the pitch. The final third has universally been noted as an important area of the pitch and real danger can threaten once teams being to take control of this portion of the playing area. City are again above the norm, completing 106 of 145 such passes, where only 100 were predicted compared to Bolton's 66 from 112 and an expectation of 79. If representative over a league wide controlled dataset we see the double whammy that lesser sides face up to against the best. City make more passes from the final third than their opponents and because they have more talented players, they are also better at such passes.

Equally instructive is the record of passes made inside a line drawn parallel to the edge of the box. City made over 50 such attempted passes, completing 34 compared to an average team's expectation of 29. Bolton could only attempt 24 such tries, completing only 6 against a norm of 10.

Reo-Coker stamps his authority on the midfield.
Needless to say one game is insufficient sample size to define a player's passing abilities, but we can use similar pass expectation models to quantify how individuals performed on that particular match day and maybe hint at their overall ability. Nigel Reo-Coker anchored the Bolton midfield and captained the side over the season and is currently a free agent looking likely to continue his career in the Championship.

Below I've screen grabbed his 22 first half pass attempts from the home game with Manchester City. I've chosen the first half merely for clarity of picture. The screen grab illustrates the direction and difficulty of the passes he attempted and in the table below I've referenced each pass along with a brief description, the outcome, (1 for a success, 0 for a failure) and the expected completion rate for such passes derived from all the passes attempted by City and Bolton in the game but minus Reo-Coker's 22 first half efforts.

The cumulative completion rate expectancy was that Reo-Coker would complete 17 of his 22 passes and that's exactly the figure that he was successful with. Again if we make the not inconsiderable assumption that a combination of City and Bolton passes equate the EPL average, this would make the player an average EPL passer of the ball, but above average within that Bolton side, again consistent with his current career stage that flirted with minor international honours, but looks set to continue at a lower level.

Nigel Reo-Coker's 22 First Half Passes At Home To Manchester City, EPL 2011/12.

Pass no. Completion. Expectation%.
1 1 82 Forward pass towards halfway.
2 1 98 Back pass to keeper.
3 1 88 Forward pass from own area.
4 1 94 Short, forward pass to flank from own half.
5 1 80 Forward pass to flank from centre circle.
6 1 94 Square ball from inside own half.
7 1 83 Short pass inside the centre circle.
8 1 80 Attacking pass to final third from halfway.
9 1 74 Square ball to central area in final 3rd.
10 0 10 Attacking pass to area from own half.
11 0 83 Forward pass towards halfway.
12 1 70 Pass from own box towards halfway.
13 0 42 Diagonal pass from halfway to area.
14 0 66 Forward pass to flank from own half.
15 1 90 Short, forward pass inside own half.
16 0 87 Short, forward pass inside centre circle.
17 1 67 Pass from final 3rd to flank. Goal buildup.
18 1 96 Backward pass from halfway.
19 1 88 Diagonal pass to flank from own half.
20 1 85 Diagonal pass to flank from own half.
21 1 84 Backward pass from inside centre circle.
22 1 88 Short forward pass from own half.

Again I'm reluctant to make any bold claims based on just 1000 passes from a single game, but some self evident features of the modern game appear to be confirmed by the various regressions. Passing the ball backwards, especially from inside your own half is a very low risk action, no doubt in most part because players will only choose to make such passes if they are supremely confident that they will retain possession.

Passing the ball infield from the flanks carries a greater risk of an incompletion, partly due to the touchline restricting choice and allowing tacklers to jump the pass and lastly passing the ball forward, especially from the final third into central areas of the pitch becomes much more difficult, again for obvious reasons.

Reo-Coker's passing timeline illustrates some of these overall features. Each of his incompletions came on attempted forward passes, three of which were among the longest he made and therefore also carried the largest chance of failing to reach their intended target. The vast majority of passes had a completion expectation of 80% or above indicating that player passing evaluation certainly requires an input that considers the difficulty of the passes attempted rather than being based solely around completion percentages. Had Reo-Coker been less adventurous he would have inflated his completion rate, but as pass 17 illustrates he would likely have hurt his team overall.

Reo-Coker's most noteworthy pass of the half was number 17, not only did it carry a relatively high level of risk, it was made from around the final third to the flanks, traveled a fair distance and it set up Petrov to supply the cross for Klasnic to score Bolton's opening goal, allowing the hosts to remain competitive throughout the match.

1 comment:

  1. That was quick, yes pass 7 & 14 are transposed on the graphic :-). thanks to anon for the heads up.