Season Review: Goalkeepers

Written by: Christian Wulff (@ahellofabeating)

One of the more interesting developments within the football analytics scene the last couple of years has been the increasing amount of advanced statistics trying to judge the performance of goalkeepers. This makes sense: football is in many ways a lot more difficult sport to analyse with data then the more ‘rigid’ type of games prevalent in the home of sports analytics, the USA. The size of the pitch, the amount of players, the flexibility of who, how and when you can attack and defend, all make it a lot more complex game to quantify using numbers. The goalkeepers’ specialist role and defined area helps limit the amount of set actions they are supposed to fulfil, making them (a little bit) easier to analyse with such data.

In February this year I introduced a new Expected Goals (xG) model for Premiership goalkeepers in an article for 90 Minute Cynic. If you’re interested in getting into the detail of how xG works and how I’ve applied it to measure shot-stopping abilities (and some great vintage pictures of history’s top goalkeepers) it’s worth a read.

In short,  whereas standard xG models measure the quality of a chance at the point of an attempt being taken, my model takes into account two more criteria.

  • Whether the attempt was on target (thereby forcing the goalkeeper to attempt a save)

  • The quality of the strike itself.

Where an xG model would give the probability of a goal being scored at the moment before the ball is launched towards the goal, this model will measure the likelihood of scoring from a shot that is on target, with a certain strike quality and from a specific location. Or, fitting to this article, how likely it is to being saved by the goalkeeper.

(for more detail on how Strata judge the shot quality of a strike, see the article mentioned above)

Using this adapted model the shot-stopping performances of the Premiership goalkeepers are compared against the average save rate of all the goalkeepers in our database, consisting of over 50,000 attempts on target from 15 different leagues over the last two years.

While the methodology behind the model can seem complex, the principle is simple: considering the type and quality of shots they have faced this season, how many goals would a goalkeeper be ‘expected’  to concede, based on the historic average of all the saves in we have recorded.

It’s worth mentioning that in this review we’ll only be able to judge  shot-stopping and not the many other facets of the role; dealing with crosses, distribution and organising the players in front of you. There is also the caveat that we’re only looking at shots on target, as they are per definition the attempts that require a save to be made. A goalkeeper could help influence attackers to miss the target and we’re not able to include such potential positive actions in this analysis.

Before we review the individual goalkeepers, let’s look at each Premiership club combined: how many goals have their goalies saved or cost them over the season?

When reviewing this data, it is important to note the average quality of chances the goalkeepers of each club have faced. Two goalkeepers might have faced shots of the same combined xG value (i.e. the quality and quality of each shot over a season), but if Goalkeeper A faced 75 shots and Goalkeeper B faced 100 shots, the former is actually likely to concede more goals, because of the concept of random variance (see Matt Rhein’s article for a good explanation of this in practice). This is why the average xG value of each shot faced forms a part of the analsyis.

GK Graph 1.PNG

Overall, it has been a fairly good season for the Premiership goalkeepers. In five of the clubs, the shot-stopping abilities of their numbers 1s resulted in four goals or more being prevented than what could have been expected based on the historic norm. Another three clubs performed just above the average, while there were clear issues in the remaining four clubs, its goalkeepers conceding more goals than expected.

In these results, there is an interesting link to the analysis of centre-backs I recently did for Modern Fitba. In it, I was a lot more critical of Hearts' defence than the general consensus around their performance this season. While the team conceded the third least goals in the league, in terms of the quality of chances they allowed they were only eight. Motherwell was the opposite; eight in goals against, but third in quality of chances conceded.

The explanation for why becomes a lot more clear when we look at the shot-stopping abilities of the two clubs’ goalkeepers this season. Hearts’ goalies saved a remarkable 10.5 goals more than expected, which is over four more than the next best club, Celtic. While they did face the second lowest average quality of chances (xG value of 0.28 on each shot), it is still a remarkable performance.

It is the reverse story at Motherwell, where the goalkeepers cost the club 3.5 goals over the season – if you take that away from their total goals conceded, they would have been sixth in non-penalty goals against this season, ahead of Rangers and St. Johnstone, and narrowly behind Kilmarnock and Hibs.

Dundee’s centre-backs were also heavily criticised in my centre-back review and it’s clear that they owe a lot of gratitude towards their goalkeepers this season, with the 4.5 goals saved by Elliot Parish, Scott Bain and Callum Ferrie a big contribution towards keeping them in the top-flight this season.

The numbers are not pretty for Hamilton, with 7.3 goals more than expected conceded – they also have the lowest pure save percentage, with only 62.8% of all shots on target saved. At least they faced the second highest quality of chances. There are no such excuses for the Ross County goalkeepers. Not only did they face the lowest quality of chance on average – something which should make a save more likely – they were bottom in terms of the actual shot-stopping, costing their team 8 goals over the course of the season.  

In the centre-back review, I noted that Ross Country’s defence was the biggest underachievers when it came to Expected Goals, i.e. they conceded far more goals than what could be expected from the chances they allowed.  The difference between their goals against and the expected goals value of the chances conceded was 7.5 goals, almost equal to the amount their goalkeepers underperformed by. It’s a damning statistic.

Now let’s narrow the analysis down to the individual goalkeepers, looking at all goalkeepers that played more than 1000 league minutes in the 2017/18 season.

GK Graph 2.png

There are two goalkeepers that stand out from the rest. With Hearts' excellent team result it's no surprise that Jon McLaughlin is one of them (although as you’ll see below, Jack Hamilton also made a very positive contribution in his five games).

However, McLaughlin is marginally beaten when it comes to how many goals he saved his team by Allan Manus. The Saints stopper prevented 0.26 goals more than expected per every 90 minutes he played, 0.01 more than McLaughlin. Over the course of a full season, that would mean a total of just under 10 goals. Mannus’ performance is even more impressive since he faced on average the second highest quality of chance in the league (McLaughlin faced the lowest quality).

It is quite the irony that the two goalkeepers that have produced the best shot-stopping in the Premiership this campaign will both have left their clubs before the next season starts. Zander Clark is Mannus’ likely replacement and his shot-stopping has been decent this season, although not any more than that: he’s 9th of the 15th goalkeepers with 1000 or more minutes played, saving an equivalent of only 0.6 goals more than expected over a whole season. St. Johnstone need to either reduce the quality of chances they concede (they were 4th worst in the league last season) or Clark has to replicate Mannus’ shot-stopping performance next season. Otherwise they might well end up in relegation trouble next season.

Here are the statistics in table form:

GK Table 1.PNG

After Mannus and McLaughlin there are four goalkeepers grouped closely together with Tomas Cerny in third place, just above Craig Gordon. Cerny has also faced the highest quality of chances in the league, making his performances even more impressive.

Cerny’s numbers are a good example of how this model can be a lot more nuanced than the standard save percentage often used to rate shot-stopping: Cerny actually had the 4th lowest rate when it came to the percentage of shots he saved, at 67%, over 11% less than McLaughlin who had the highest rate. Yet Cerny is only one step below the Hearts goalkeeper when it comes to this shot-stopping metric; mostly down to the fact that he was required to save a lot tougher selection of shots.

The Celtic number 1 and 2 follow in the next two places, and together with Dorus De Vries, Gordon and Bain combined to save over 6 goals more than expected over the course of the season – the second highest of any club.

Wes Foderingham shows up as a slight statistical curiosity; he saved exactly the amount of attempts expected of him this season. His numbers confirm the impression I have when watching him play, he is a perfectly serviceable goalkeeper but he is not likely to win Rangers any more games than they would have anyway.

In comparison,  Joe Lewis and the other Aberdeen goalkeepers saved over 4.5 more goals than expected over the season than their counterparts at Rangers – that could well have been one of the small differences between 2nd and 3rd place for the Ibrox club this season.

Trevor Carson and Ofir Marciano have both been getting plaudits for their displays this season. Carson especially has been seen to have had a very good season, with a call up to the Northern Ireland squad and being heavily linked to Celtic in the January transfer window.

This shot-stopping model  is not so kind to either of them. As mentioned previously, Motherwell allowed a lot less chances than their actual goals conceded would indicate. That is a good performance indicator for their defence, but the numbers suggest that this discrepancy was partly Carson’s fault: he is 11th out of the 15th goalkeepers in shot-stopping.

Marciano stats are far worse. He conceded on average 0.14 more goals per every 90 minutes played than expected, which equates to almost 5.5 over a whole season. Compared to his counterpart across the city, that’s almost a 15 goal swing between the two Edinburgh goalkeepers. Marciano also has the lowest save percentage in the league, at 64.3%. While the Israeli is virtually tied with Gary Woods in the rankings, there is then a sizeable gap to the two goalkeepers at the bottom of our table.

Unfortunately for Ross County, they both play for them. It’s been a rough season for Scott Fox and Aaron McCarey. They both conceded 0.21 goals more per every full game than expected. As noted above, with having cost County a combined 8 goals this season they are likely to have played a significant part in their relegation, a blame they share equally; there’s hardly any difference between the two both in terms of their expected saves or average quality of chances faced.

Scott Fox.jpg

There were 13 other goalkeepers appearing in the Premiership this season, all playing too few minutes to be included in this analysis. Their samples are simply way too small to draw any significant conclusions from; data over only a few games are likely to show a lot more swing in numbers, before they would normally stabilise and regress towards a mean once more minutes have been played 

That large variance is seen in the table below, especially on save percentage, which goes betweeen 20 to 100% over these small samples.

GK Table 2.PNG

Dorus de Vries is top of this table, and it does confirm the general impression of him; a more than good enough goalkeeper for Celtic in the SPFL, but one who has fallen through when tested on a higher level in Europe.

When it comes to shot-stopping, Alan Mannus and Jon McLaughlin were the stand-out performers this season. With both Hearts’ goalkeepers leaving the club, there is a lot to live up to for their new signing, Zdenek Zlamal.

Celtic as a goalkeeping collective were very solid and probably made a larger contribution to their title-winning season than many would have been aware of.

Tomas Cerny was definitively not a contributing factor to Partick Thistle’s relegation and could be an good option for several Premiership clubs next season.

Unfortunately for them, the Ross County goalkeepers came out far worse than anybody else. Compared to the Hearts goalkeepers, there was an almost 19 goal difference between the impact of the shot-stopping produced.

It was a crucial difference for both clubs this season and shows the often underrated importance of a very good shot-stopper.