Goalkeepers are weird. They break the fundamental rule of the sport in that they use their hands.
A popular joke that used to make the rounds was, “What do you call someone who hangs out with soccer players? A goalkeeper.”
Unsurprisingly soccer analytics have found it difficult to understand or quantify the enigma that is the goalkeeper. The main problem with using statistics to examine the role of a keeper is that when he or she is involved in the play it is usually because something has gone wrong.
Soccer Central podcast: SPORTSNET.CA’s Soccer Central podcast, hosted by John Molinaro and James Sharman, takes an in-depth look at the beautiful game and offers timely and thoughtful analysis on the sport’s biggest issues. To listen and subscribe to the podcast, CLICK HERE.
When a team is defending they have three main objectives. The first is the most basic, to win the ball back. The second is to prevent the opposing team from shooting—simply put, if they don’t shoot they can’t score. The third is to ensure that if the opposition does get a shot it is from a poor position with a low chance of scoring.
Keepers are involved in all three of these objectives. They come to gather crosses and loose balls, which is part of winning the ball back. They cut down angles to try and make shooting a less enticing option. Finally, if possible, they try to force the shooter into a poorer position. However, it is when the defence fails and the opposition gets a shot off that we really start to pay attention to the keeper.
This is fundamentally how we assess keepers when we are watching the game—do they save the shots fired their way?
To answer this question the obvious statistic that comes to mind is save percentage. What percentage of the shots on target does the keeper save? Save percentage, however, is very problematic when we consider the role of the defending team and how different teams prioritize the three objectives of defending.
Some teams put major emphasis on winning the ball back—Barcelona under Pep Guardiola made it their priority to win the ball back within five seconds of conceding possession. Other teams minimize the number of shots they concede by playing a high defensive line and trying to catch opposing forwards offside. This strategy of playing an “offside trap” prioritizes lowering the number of shots the opposition takes while accepting that the quality of shots will likely be quite high.
Tottenham under Andre Villas-Boas played a high defensive line and as a result gave up very few shots, but the ones they did give up were often breakaways or fast breaks.
Now consider the role of the Tottenham keeper Hugo Lloris. The Frenchman didn’t face a lot of shots when playing for Tottenham under Villas-Boas because the high line caught a lot of opposing strikers offside. However, the shots he did face were of a very high quality and he had a below league average save percentage. Was this a reflection of Lloris’ ability as a goalkeeper? Of course not—it was a reflection of the system.
This is just one example of how save percentage, when used to evaluate a keeper, can be unreliable, but there are many more. Sometimes teams get lucky and they go through a stretch where every striker they face fluffs their chances straight into the goalkeeper’s arms. Other times teams go through an unlucky stretch where all of the strikers they are facing are clinical and find the bottom corner with every shot. These are factors that may have nothing to do with the defenders or the goalkeeper, yet they all play into save percentage.
Given all of the factors it is no surprise there is almost no consistency between a keeper’s save percentage from one season to the next.
The following graph features the 10 goalkeepers who played the majority of games for their teams in the Premier League last campaign and this season. The graph shows no relationship between save percentage from one season to the next. It is also interesting to note this season playing under a different system at Tottenham that Lloris’ save percentage has increased dramatically.
Analyst James Grayson has written about many more of the problems with save percentage, but the basic message is the same: we know some keepers are better than others, so if save percentage varies so much from year to year it can’t be a good way of determining which keepers are good and which aren’t.
Another method to compare keepers was developed by Colin Trainor who looked at using a special Expected Goals model for keepers. The idea of this model is to calculate the chance of a shot going in based on from where the shot was taken, the type of shot (header, right foot, left foot) and where it landed (top corner, center of the goal, etc). The keeper is then evaluated based on how many goals he conceded compared to how many goals the average keeper would have conceded based on the expected goals model.
This model performs better than raw save percentage, but still has some flaws. It doesn’t take into account where the keeper’s starting position is or where the defenders are relative to the shooter. These are factors related to the three objectives of a defending team mentioned earlier.
Even this more advanced technique still only tells us how keepers deal with shots. There is much more to being a goalkeeper than just stopping shots. Keepers must organize their backline, offer a support option to defenders and often play as a sweeper themselves.
To develop a more satisfying tool to evaluate keepers we must take into account how the team prioritizes the three objectives of defending, how well the opposing team is testing the keeper with their shots and how the keeper contributes to his team beyond shot stopping.
Until we come up with some statistic that takes these factors into account we’re left with a model that does a better job than save percentage, but is still incomplete.
Even in the age of analytics the goalkeeper remains an outlier and an oddity.
Sam Gregory is soccer analytics writer based in Montreal. Follow him on Twitter