This post has been written by Dan Kennett.
In the last year, this site has led the way on Goalkeeping analytics. In that time the focus has become more on “Expected Saves” rather than the humble old Save% (Sv%) i.e. [Total Number of Saves] / [Total Shots On Target Faced]
Following a chance discovery on the NBC Sports website, it’s been possible to quickly collect Sv% data going back to 2007/08 for England & Germany and 2008/09 for Italy & Spain, resulting in almost 48,000 Shots On Target with an average of 360 saves for 95 Goalkeepers (Individual Goalkeeper sample size is currently the main constraint with Goalkeeping models).
With this data it’s now possible to re-revisit the humble old Sv% and ask if some Goalkeepers are betting at saving shots than others over the course of their career. All Goalkeepers with > 100 saves were put into a funnel plot and the results are below:
The purpose of the funnel plot is to show how randomness decreases as the sample size increases. In this case, as a Goalkeeper faces more Shots On Target. This is represented by the curved lines getting closer to the horizontal line.
- The horizontal black line is the overall Sv% for the 48,000 shots (72.13%)
- The curved black lines represent 2 standard errors above and below the mean
At the top centre of the chart there are 5 dots above the curved line that tally with the received wisdom of “good goalkeepers”: Gigi Buffon, Marc-Andre Ter Stegen, Petr Cech, Christian Abbiati and Manuel Neuer. Maybe Neuer genuinely does deserve his unofficial title of “world’s best Goalkeeper”?
Some love should also be shown for Tim Howard who has been consistently excellent for Everton over 7 seasons.
At the bottom centre of the chart there’s a cluster of 3 keepers from the Premier League who should set alarm bells ringing for fans: Brad Guzan, Boaz Myhill and Tim Krul. This statto will also be keeping close tabs on Ruben Ivan Martinez of Rayo Vallecano from now on!
Sv% is normally distributed (p = 0.256)
There is a weak relationship between Sv% and SoT Faced (r = 0.345) but the p-value is very low (0.001)
We can rule out “league effects” as once goalkeepers from different leagues are grouped together and compared, there is no significant difference amongst the means (p=0.471)
The following is a copy of the collated data: