Goalkeepers Season Review Part 3: Distribution

The third and last part of Modern Fitba’s season review of goalkeepers in the Premiership focuses on a part of the role that is only increasing in importance and one that some managers are now valuing as much as saving shots: distribution.

It is also the part of the review where it is most difficult to determine a clear ‘winner’. The shot-stopping analysis had a clear objective (who makes more saves that what could be expected from the shots faced), and while it’s trickier to rank goalies based on their dealing with crosses, we could measure what is objectively an important skill: the ability to claim high crosses into the area, stopping a potential dangerous situation and regaining possession.


I Like Your Style

The main issue with analysing a goalkeeper’s distribution through data is that so much of the output will be directly influenced by a team’s playing style and tactics. A keeper for a superior quality team that prefer to build up attacks through controlled possession is likely to have a high pass completion rate as they’ll be instructed to play short out often to defenders in plenty of space and time, their opponents already having retreated back to their own half.  A goalie for a team that place a lot more importance on being direct, getting the ball as far up as quickly as possible can expect a far lower percentage of their passes to be completed, having been told to ‘go long’ into areas filled with opposition players.

It means that analysing a goalkeepers’ distribution will be as much about the style as it is about the quality. Earlier this year Seth Dobson brought the concept of Pass Sonars to the Scottish Premiership – it’s a perfect starting point when looking at any player’s passing profile as it captures the direction, volume and length of all passes attempted in one picture:

GK Sonars.png

The majority of the Premiership goalkeepers covered in this season review have a very identical passing profile with the vast majority of passes going straight forward, all with an average length over 30 meters. The three goalies that stand out from this standard are – to varying degrees – Scott Bain, Craig Gordon and Allan McGregor.  But there are still significant differences between them, even when comparing the two Celtic goalkeepers, with Bain going far less long and with many more passes out wide than Gordon.

Otherwise, it is clear what instructions most Premiership goalkeepers have been given: go long and straight ahead.

But how can we start unpicking the quality of passing?

 xPassing

As touched upon earlier, a goalkeeper’s pass completion rate will be largely determined by the options they are given and what is expected of them when they have the ball. Perhaps unsurprisingly, the distance of a pass is the single most important factor in whether the pass will be completed or not:

Graph shows the average completion rate for all passes in the Premiership last season, based on the length of the attempted pass.

Graph shows the average completion rate for all passes in the Premiership last season, based on the length of the attempted pass.

Other than for very short passes (which is often recorded as short because the pass is intercepted or blocked up close), the clear general rule is that the longer the pass, the less chance it has of being completed.

Another issue with looking solely at pass completion rate as a metric to determine the quality of passing is that it tends to ‘reward’ safe passes, often sideways and backwards, rather than more purposeful forward passes, that also carries a higher risk of not being completed. As an example, the top 5 players for pass completion (among those who played at least 3 full games) in the Premiership last season are all Celtic centre-backs. Number 6 is Nir Bitton, who has also spent time in that position.

For other positions on the field there are metrics through which we can better judge a player’s passing ability, such as looking at the quality (Key Passes) and quality (Expected Assists, i.e. xA) of the chances directly set up, how many attacks players were a part of that resulted in a chance and the quality of that chance (xG Build-up) and also how many times they moved the ball into the most dangerous attacking areas on the pitch (Deep Progressions).

There is also a more all-encompassing metric being developed that is a much better alternative to pure pass completion rate and which is more relevant to goalkeepers: Expected Passing (xP)

Matt Rhein introduced this new passing model for the Premiership earlier this year and it works on the same principle as its more famous cousin Expected Goals (xG): based on the location and end point of each pass, how many passes are on average completed? Or in other words, how many of those passes can we ‘expect’ to be completed.

The model we have for the Premiership is in its early stages and we’ll be able to add more detail and make it more robust with more seasons’ worth of data, but it already creates an alternative metric to pass completion rate as it takes into account the presumed difficulty of each pass.

xPassing List.PNG

The model is based on passes by players in all positions so it’s perhaps not surprising that most of our goalkeepers complete less passes per every 90 minutes played (p90) than can be expected: they are goalies after all! And again, the result will also be driven by the length of a pass, even if the difficulty is part of the model: if you attempt more difficult passes (i.e. longer) the variance of the result will be larger (for more detail on this principle see Danny Page’s excellent article on variation and expected goals.)

Scott Bain tops the xPassing table, making 2.2 more passes than expected p90. That’s almost a full pass more than Craig Gordon and 2 more than Allan McGregor. With so many caveats to the rating, especially the influence of team playing style, some of the most interesting findings can be found in analysing the output of keepers from the same team. Most of these pairs are grouped quite closely together, but in addition to Bain’s clear difference to Gordon, Vaclav Hladky is producing significant better passing numbers than his St. Mirren colleague Craig Sampson (we’ll return to the possible reasons why) and there is also a difference of more than 1 pass p90 completed between the goalkeepers at Kilmarnock and Motherwell.

While the xPassing ranking is quite similar to pass completion rates for kicked passes, the Kilmarnock example is a good illustration of where Expected Passing gives a different result:: Jamie MacDonald’s completion rate is 1.2% lower than Daniel Bachmann but he has completed 1.2 more passes p90 than expected. 

xPasing + Completion table.PNG

Go the Distance

As mentioned previously, the length of a pass is the major factor in determining whether a goalkeeper will complete it successfully or not.

The below graph compares the completion rate to the average distance of each goalkeeper’s kicked passes (this excludes passes from goal-kicks and free-kicks). Scott Bain and Craig Gordon attempted the shortest passes on average (27.3 and 29.5 meters) and have the highest completion rates. Trevor Carson has the lowest % of passes made and the longest average distance attempted (53.4m).

Kicked Passes Complete + Length.PNG

 It’s a similar trend when we look solely at goal-kicks:

GoalKicks graph.PNG

Goal-kicks are especially interesting in this regard as it’s a clear indication of a team’s preferred style: are they happy to push the team up and have the goalie kick into a 50/50 duel high up on the pitch, or do they want their defence to drop deep and pick up the ball. There is quite the difference between Celtic and Rangers in this regard with Bain’s and Gordon’s combined average of 38 meters a whole 11 meters shorter than the goal-kicks Allan McGregor took last season (in the table below you can see it’s a trend replicated both for kicked passes and free-kicks). In fact, McGregor’s goal kicks were actually just shorter than those of St. Mirren’s Vaclav Hladky.

There is a dramatic shift in the length of kicked passes, goal-kicks and free-kicks after Hladky replaced Craig Samson. While Samson had the 4th longest passes on average, Hladky had the 4th shortest last season (Hladky actually has the second longest thrown passes, and the third most thrown passes – it’s an option he often use). There was a similar trend at Motherwell where Mark Gillespie went on average around 8 meters shorter than Trevor Carson, both for kicked passes and goal-kicks.

Avg Distance List.PNG

Split the Difference

Looking at the average distance of each pass will give big clues to a goalkeepers distribution choices, but some of the finer detail might be lost. Below is all the goalkeepers’ kicked passes split into four groups: Short (0 to 14 meters), Medium (15-29m), Long (30-52.5) and Very Long (52.5m+). The last category is based on the average permitted football field length being 105 meters (all pitched must be between 90m and 120m), so a very long pass is one longer than one half of the full pitch.

Distance Grouping Shortest.PNG
Distance Grouping Longest.PNG

A few interesting points are highlighted: Allan McGregor’s average pass is the third shortest in the league, but he’s only got the 5th most short passes (and only 1.3% more than Zdenek Zlamal in 7th). McGregor has also got the 6th highest % of long passes. It’s clear that he’s most comfortable in that middle range of passing between 15 and 50 meters.

Jack Hamilton has the 6th highest % total in the Very Long category but he is also 3rd highest in the Short pass category – his profile is the opposite of McGregor as he usually either goes short or very long.

Trevor Carson has the longest average pass and it’s no wonder why when you look at the distance split of his distribution: only 9.6% of all his attempted passes last season were less than 30 meters long.

Distance Split Passes Table.PNG

Take Your Time

The last angle we’ll approach this distribution analysis from is looking at the decisions goalkeepers make on whether to try to kick the ball immediately upon receiving it or whether they decide to take a touch before attempting a pass. This will of course be situational, with no one correct answer, but it’s an interesting additional layer to the analysis. In the graph below the % of the goalkeepers kicked passes that were direct (attempted without taking one or more touches first) is compared to the average time between them receiving a pass until they attempted a pass themselves.

Direct Passes vs Time.PNG

 Interestingly it is Craig Gordon who tops the league in the % of direct kicked passes: almost 1 in 4 of the passes he attempted was made without a first touch. Adam Bogdan spends the least amount of time before each of his passes, 2.9 seconds to be exact, which is 1.2 seconds more than Jamie McDonald who is the ‘slowest’ passer.

In general, it is goalkeepers from teams that are happy to spend some time getting their team higher up the pitch in preparation for a long ball (St. Mirren, Motherwell, Hamilton) that take the longest time on the ball and have the fewest direct passes.  Teams that in general want the ball in play faster and for passing to be quicker – Celtic, Hibs, Rangers – have goalkeepers that spend less time on ball and have more direct passes.

There are some outliers to this: Liam Kelly has the third fewest direct passes but is 5th quickest on the ball while Zander Clark has the 6th highest direct passes but is also 6th when it comes to longest time spent with his ball at feet.

Conclusion:

As stated in the introduction, it’s difficult to rank goalkeepers from ‘best’ to ‘worst’ when it comes to distribution, as so much of the data output will be influenced by team style and the instructions given to the keepers. The best indicator for pure ‘pass quality’ is xPassing, in which Scott Bain – perhaps surprisingly – comes top. But that Craig Gordon comes second is likely to be an indicator that even this metric is heavily affected by the type of passes a goalkeeper is asked to perform within their team: few would argue that Gordon is the second best passer of the ball among Premiership goalkeepers.

So the most interesting analysis can perhaps be found when comparing players within the same team. Scott Bain’s stats are significant better than Gordon’s which does ‘pass the eye test’ when it comes to watching Celtic games.

Vaclav Hladky’s xPassing stats are significant better than Craig Samson’s and their individual passing style breakdown is also quite different. Is Hladky simply a more comfortable passer of the ball? Is this confidence influencing his decision making which makes him go shorter, or are the two goalkeepers simply reacting to different tactical circumstances?

Such questions for the St. Mirren – and other clubs’ – goalkeepers can be further explored to detailed video analysis.

This distribution analysis is the third and final part of the Modern Fitba 2018/19 season review for goalkeepers. Throughout 2019/20 we will continue to analyse the advanced data to monitor how well – and in what ways  - the Premiership goalies are performing their duties.