Goals Saved Above Average: A look at the best (& worst) goalkeepers in the Scottish Premiership?
written by: @TheGersReport
For anyone who has ever published a blog, or even sent out a tweet, sometimes you can’t help yourself…you can’t stop yourself from checking out how many ‘views’ your work has gotten. Even though I haven’t published anything on my old site in over a year, I still check regulary to see what old posts are still getting read. Interestingly, the one post that gets checked out on a near regular basis is one that was a simple table of advanced goalkeeping stats for Scottish Championship goalies from the 2015-16 season.
Every other week or so, there’s a little run of hits on that post & it serves as a reminder that…I used to really enjoy keeping track of goalie shot-stopping stats. Now, I freely acknowledge that shot-stopping is only one isolated aspect of a goalie’s job & there are much smarter people than me out there adding many more layers to the statistical analysis of goalkeepers…but there is still a lot that can be learned from shot-stopping stats.
Like any dabbling in fitba analytics…it’s with the outliers that the real learning can happen…like this example from 2017 that highlighted how analytics could prevent poor recruitment of squad players.
This time around, I wanted to continue my trend of simply lifting good ideas from the hockey analytics world to see what it would look like when applied to the Scottish Premiership.
One of the most embraced advanced stats for evaluating NHL goaltenders is Goals Saved Above Average, which takes a look at a goalies save percentage numbers & compares it to the league average for the kinds of shots that goalie faced. Basically, if you put a league average goalie in the exact same situations as the goalie being evaluated….how many more (or less) goals would that league average goalie allow.
For example, the goalie with the best Goals Saved Above Average (GSAA) rate last season in the Premiership was Hibs’ goalkeeper Ofir Marciano. If the league average goalie came in & faced the exact same shots on target that Marciano faced…that league average goalie would have allowed 4.23 more goals. On the flip side, the goalie with the worst GSAA rate was Hamilton’s Gary Woods. If the Accies threw the #1 jersey on a league average goalie in Wood’s place, that league average goalie would have allowed 15.67 LESS goals than Woods did.
Of course, we are dealing with statistical hypotheticals here & the goal of the descriptions above is to help people understand what the statistic represents & not to be taken super-literally.
Below you’ll find the complete table of every goalie who was forced to make at least 20 saves last season & their Goals Saved Above Average rates.
For the most part these results pass the sniff-test. Goalies like Marciano, Daniel Bachman, Liam Kelly & Scott Bain were spoken of in pretty positive ways last season. Bachman has been linked to Rangers, Kelly moved down south to join Queens Park Rangers, & Scott Bain went from being Neil McCann’s least favorite goalie to supplanting Craig Gordon for club & country.
Probably, the biggest surprises here that differ from the mainstream narrative may be the fact that Aberdeen’s Joe Lewis & Rangers’ Allan McGregor are in the negative here. For Lewis, the numbers have been telling us for a while now…that he may not be that good. Then there’s McGregor, who made most people’s team of the season…& whose timely saves truly seemed to save Rangers consistently in 2018-19.
We’ll dig deeper on both of these goalies in a little bit but first…..
As you know, when doing stats based analysis, you typically like to break numbers down to a per 90 minute rate to essentially even the playing field when looking at counting stats. The tricky thing with goalies is that a goalie playing for a team like Dundee has a whole lot more to do in a 90 minute match than a keeper from Celtic does.
So, rather than breaking down to a 90 minute rate…I want to break down Goals Saved Above Average to a per 3.45 Shots on Target rate. Why 3.45? On average, goalies faced 3.45 Shots on Target per match last season. Of course, it may take Scott Bain a match & a half to get there & maybe even only a half for someone like Dundee’s Jack Hamilton…remember the goal is to account for an unequal distribution in minutes for the list above.
Again, facing 20 Shots on Target is the cut-off for the table below…
This would be my go-to number when answering the question: “Based on the numbers, who was the best shot stopper in the league last season?”
Daniel Bachmann, who has since returned to Watford, was at a whole other level last season. His overall Save Percentage was 0.789 (the third best rate in the league), despite the fact his Expected Save Percentage was 0.742 (the 13th highest rate in the league). For a comparison, Scott Bain led the league with a ridiculous 0.909 Sv%…which was partly driven by the fact his xSv% was 0.888 (also the highest in the league). Bachmann had much more difficult saves to make, which highlights the value he brought to Killie last season.
At this point, I think it’s best to show how this kind of goalkeeping data can be parsed out to create entry points for performance analysts.
Another aspect of hockey analytics I have “borrowed” is the concept of categorizing shots by how ‘dangerous’ they are…basically this is a way to separate low percentage shots which rarely score from shots that tend to have a high success rate.
The concept of labeling these ‘danger zones’ to easily differentiate shot quality came from the two analysts who ran the WAR on ice blog & who were soon both promptly hired by NHL teams (the Minnesota Wild & Pittsburgh Penguins).
For the application to football, I have labeled the following shot locations:
Shots beyond the arc at the top of the box are Very Low Danger Shots
Shots from the arc to the edge of the box are Low Danger Shots
Shots inside the box, but on the edges (non-shaded area in visual above) are Medium Danger Shots
High Danger Shots are in that shaded area from top of box to edge of six yard box
Very High Danger Shots come from within the six yard box
Headed shots have the same labels but those from the High Danger zone aren’t really ‘high percentage’ shots but for consistency I’ve kept the label.
These are the shot location labels I used to organize the data for the Shots on Target that goalies faced. Below you’ll find the league average save percentage from each category of shot.
Right away you can see why it’s imperative to make the distinction from all kicked shots inside the box & shots that come from a more central area. The difference between the Save Percentage on Medium Danger shots & High Danger shots is pretty extreme. This is why I really value playmakers who can get the ball into that area. You can also see the huge difference between headed shots from the heart of the box & inside the six yard box.
Now remember, we are looking to see how this kind of data can become entry points for performance analysts to find ways to improve both the play of their goalkeepers & the play of the defence in front of them.
The infographics below highlight some key stats for a goalie, along with what percentage of the Shots on Target & of goals allowed that come from each danger zone. To add context, I compared these percentages to the league average rates. For example, 18.4% of the Shots on Target that Daniel Bachmann faced were Medium Danger kicked shots…which is equal to the league rate. On average, 12.1% of the goals allowed by Scottish Premiership keepers came from this same zone, for Bachmann his rate was 0%.
Now, you can see how that is represented below:
When examining the results for Bachmann, the clear outlier begins with the fact that 31.6% of the saves he had to make were High Danger shots (kicked), slightly higher than the league average of 25.9%. His Save Percentage on these shots was 0.458 which is below the league average. This led to 81% of the goals against coming from High Danger shots. League-wide, it’s basically a coin flip on whether a shot from this zone will be saved or not. Which means, a major focus for Kilmarnock should be on how to limit the chances they allow from this shot location.
Beyond those High Danger shots, Bachmann was basically unbeatable…posting a 0.942 Sv% on all other shots.
I’m sure many of you were surprised to see Dundee’s Jack Hamilton with the second best Goals Saved Above Average rate per 3.45 Shots on Target. The 25-year old had the 13th best Save Percentage (out of 21 keepers who faced at least 20 Shots on Target)…which highlights why simply using Sv% to evaluate keepers is shortsighted.
The poor defensive play in front of Hamilton meant that his Expected Shooting Percentage was the second worst in the league. For context, the worst was the xSh% of 0.653 for teammate Elliot Parrish. Parrish’s actual Sv% was 0.500…Hamilton’s was 0.692.
You can see that the two biggest outliers here in which Hamilton gave up more goals than normal both came from shots within the six yard box. This isn’t a commentary on Hamilton, it’s an indictment of the defence he played behind. You can also see that the goals allowed from High Danger shots (kicked) was way below the league average. Twenty-nine percent of the goals Hamilton allowed came from this area which compares favorably to the league average rate of 39%. This was driven by a High Danger Save Percentage of 0.636. That was the sixth best rate in the league…but the very best among the eleven goalies who faced at least 20 High Danger Shots on Target.
Like we mentioned for Jack Hamilton, allowing goals from inside the six-yard box usually isn’t on the keeper. Only three keepers allowed more goals from Very High Danger shots (kicked) than Marciano…two of them played for Dundee & the other was Liam Kelly.
Like Hamilton, Marciano’s success was driven by his ability to stop High Danger shots. Marciano’s High Danger Save Percentage of 0.750 was the best in the league for keepers who had at least ten of these saves to make. The next step for a performance analyst would be to assess what Marciano was doing to make these saves happen as a means of highlighting what works for him.
Liam Kelly leveraged his strong spell with Livingston into a move south, reuniting with Mark Warburton at Queens Park Rangers. Half of the goals that Kelly allowed came from High Danger (kicked) shots, compared to the league average rate of 40%. This isn’t a major concern because his High Danger Save Percentage of 0.543 was actually better than the league average. It really was Kelly’s excellent shot stopping on lower percentage shots that drove that High Danger goal rate up.
The main area of analysis really would have to be why he’s allowing goals from Headed High Danger shots (no keeper allowed more) & his Save Percentage was only 0.667 on these shots (remember the league average is 0.778). Is this a positioning thing? Is he not aggressive enough on crosses? That’s where a performance analyst would need to step into to identify what the issues were.
Another goalie moving south is Gary Woods who left Hamilton this summer & was signed up by Oldham Athletic. Remember, this is the same Gary Woods who had the worst Goals Saved Above Average rate in the league.
At an initial glance at the data above, you may be wondering how & why was he so bad? It has to go beyond allowing a high rate of his goals on Low Danger shots, right?
In this case, we’re best served by simply honing in on the fact that Woods’ Expected Save Percentage was 0.770 (which was actually higher than the likes of Craig Gordon, Liam Kelly, & Allan McGregor) & look at how it compared to his actual Save Percentage of 0.595 (only Craig Samson, who is now a coach, & Elliot Parish, who was signed by St Johnstone to “compete” with Zander Clark) had worse save rates.
When we break down Woods’ Save Percentage by zones…we’ll learn just how bad he was.
The question for Oldham will now be why Woods was so bad. What really hurt his save rate was those woeful Save Percentages on shots from outside the box, along with the fact that he got beat so often on those Medium Danger shots…which should be saved 80% of the time.
The last two goalies we’ll look at in which the stats don’t align with common public perception. In fact, Joe Lewis was actually named Aberdeen’s player of the year by the club for last season.
So, why was he near the bottom in Goals Saved Above Average last season? It really seems to come down to those High Danger shots. His Save Percentage on these shots was only 0.484, which was only seventh best of the eleven keepers who had to make at least 20 High Danger saves.
But that doesn’t really explain his poor overall numbers. Remember that Daniel Bachmann had a similar issue (& a worse Sv% on High Danger shots). But Bachmann’s Save Percentage on all other shots was 0.942…Lewis’ was a much lower 0.754. Interestingly, their Expected Save Percentages on all other shots was actually very similar (79% to 74%). Bachmann saved just about everything else while Lewis was basically slightly below average on the other shots he faced.
The 31-year old Lewis recently signed a five-year contract extension & a main priority for Aberdeen’s internal analysis should hone in on what happened on those High Danger shots. Are their fixable mistakes in Lewis’ approach to these chances or is it more of a systematic fix needed in the defence in front of him?
This reminds me of the Liam Kelly situation. The biggest outlier is the fact that a high rate of McGregor’s goals allowed come from High Danger shots…despite the fact that their Save Percentage on these shots is actually better than the league average.
If we take those High Danger shots out of the equation (like we just did for Joe Lewis) it’s interesting to note that McGregor’s Sv% on all other shots was 0.786……his xSv% was 0.787 on the same collection of shots.
Let’s add some context here….McGregor is 37 years old & is clearly keeping up with his younger peers. His ability to pull off the spectacular, match saving save was on full display last season & there were times in Europe where he almost single-handedly kept Rangers alive. It was these saves & performances that pushed the narrative that he was ‘the best keeper in the league’ last season. He really does have the gift of the “timely save.”
The concern is that the underlying numbers suggest he was a league average goalie & at 37, could he follow the same storyline that his former Scotland teammate, Craig Gordon, experienced last season as a 36-year old? The downward trajectory of the aging curve happens for goalies later in their careers…but it still happens...to everyone. Age is undefeated.
One last side note before the actual closing side notes section….remember that McGregor’s Goals Saved Above Average for the season was -1.01. That suggests that a league average goalie would have allowed one less goal. It’s a statistical generalization…& shouldn’t be taken too literally.
However, if he had made the save on a Jordan Jones goal back in January…a shot that was very savable…his GSAA would have been -0.01 (literally a goal difference) & his actual Sv% would have been 0.731 (much closer to his xSv% of 0.745).
Goal begins at 1 minute 9 seconds.
Stats are courtesy of data provided by Ortec Sports.
The next step in this process would be to see how repeatable these numbers are. I’m in the midst of finishing up my annual project looking at the best prospects in Scottish football. Once that’s done, I’ll go back & see if I can pull the same kinds of numbers in this post from the past season or two.
Again, all of this only looks at shot stopping. There is so much more to being a goalie but ultimately they are paid to make saves.
I used this Blueshirts Breakaway post from Drew Way as my go-to reference on goalie stats for this post & is where I developed the idea of breaking GSAA to a per 3.45 Shots on Target approach, rather than a basic per 90 rate.
Comparing the rate of Shots on Target & goals allowed in relation to the rest of the league originated in an email discourse with Matt Cane when I was trying writing about hockey. That was back in 2015…four years later I applied it to this blog post & Matt…well, he became the head of analytics for the New Jersey Devils.
This was written under the influence of the Troggs, the Raveonettes & the Jesus & Mary Chain