What were Celtic thinking? Spending £4.5 million on any keeper was an interesting move, but what could they have possibly seen in Vasilis Barkas?

I shared this Twitter thread in early August 2021, where I did a simple normal distribution calculation using Wyscout’s post-shot xG metric, and compared Barkas with Joe Hart, Fraser Forster and David Marshall. As a reminder, Wyscout data begins at the start of the 2015-2016 season. In a prior thread, I warned that adjusting to the quality of opponents would be preferable, as Hart and Forster played in much of their samples in the EPL and Serie A.

My analysis at the time of Barkas’ signing was that his distribution suggested a fairly consistent keeper, but one who did not offer ‘match winning’ shot-stopping. The reporting and scouting I read at the time indicated he was relatively strong with his feet and mobile. As has been a consistent theme with Celtic in recent years, that profile did not end up matching up with how Neil Lennon appeared to want his keeper to play.

Subsequent to that cursory analysis with very limited data, I have gained access to Statsbomb’s relative treasure trove of keeper-related data and metrics. Here is a radar comparing Hart so far this season in league games with Barkas in 2020-2021.

Celtic Way:

This radar focuses upon shot stopping-related metrics and some which have been included as proxies for ‘sweeper keeper’ play. For example, the Goalkeeper Aggressive Distance metric measures on average how far from the goal line a keeper moves to perform a defensive action. We can see that Hart has been far more forward under Postecoglou than Barkas was under Lennon/Kennedy. The number of passes from open play have also been quite different, with Hart playing more as a sweeper.

Perhaps unsurprisingly, Hart’s shot stopping-related metrics have been far superior. In fact, the key Shot Stopping % metric, which is the difference between Expected Save % and Save %, showed that Barkas was among the worst in the league last season. Yet, a single metric in the radar had me perplexed and asking the following question – then why have Goals Conceded (which exclude penalties) been nearly identical?

This metric was even more confounding to me given the relative context of team performance levels.

Celtic Way:

We can see from this table that Celtic have been performing far better as a team this season compared to last by non-penalty expected goals (npxG), with both samples limited to only league games in which each keeper played. Both statistically and likely by the ‘eye test’ of many, Celtic’s defensive organization and performance levels have been greatly improved compared to last season.

So, if the team has been performing better overall, and Hart has clearly been the better shot-stopper, then how could goals conceded be nearly identical? This is where the depth and quality of Statsbomb’s data offered a path to insight.

These two graphs showed the distribution of shots against each keeper by xG buckets (excluding penalties):

Celtic Way:

Celtic Way:

There was an important difference in shot distributions. For example, in the two buckets between 0.10 and 0.20 xG, there have been 10 shots versus Hart compared with 21 versus Barkas. We can also see the couple of huge chances against Barkas which resulted in goals. Here is a graphic I made combining the first four buckets into 0.10 increments for only chances from open play, and showing the number of shots in each bucket as a percentage of total shots faced from open play:

Celtic Way:

Next, we can break down all the shots in various ways to try and ascertain more information. Here were the distributions for only shots which were on target:

Celtic Way:

Celtic Way:

This is where the relative disparity in narratives between the two keepers appears to ‘live.’ If it seemed like Barkas hardly made saves, well that was kind of true. He faced just 22 shots on target and conceded 7 goals! In fairness to Barkas, we can again see that two of the goals were extremely high-quality chances his defence conceded.

But the story does not end there. Go back to the initial radar and take a look at the All Shots Faced and Shots Faced metrics, with the latter being shots on target. Given how poor the defence was in front of Barkas on a relative basis, how could Total Shots Faced be similar, with Shots Faced a minuscule 1.40?

How does one measure a shot that does not become a shot? Statsbomb offers significant data on balls that enter keepers’ penalty areas, and model them as ‘Claims.’ Here was Barkas’ claims heat map for last season:

Celtic Way:

As we saw once again in Sunday’s game versus St. Johnstone, Barkas is a relatively mobile keeper. This report showed his hot spots of claimed balls was pretty significant. Below the heatmap, the report breaks out each zone by claims versus claimables, with Barkas’ metric compared to league average. It shows that Barkas’ ability to claim balls was consistently better than average, and to a significant degree. How about Hart this season?

Celtic Way:

In contrast, we can see how limited Hart’s claiming actions have been per his heatmap, and he has consistently been poor relative to league average across the various zones, with the closest-central the only exception. Why does this matter?

Celtic Way:

Celtic Way:

Those two distributions show only shots coming off of crosses. Barkas’ distribution skewed towards lower quality chances, which would make sense given that he actively claimed balls in more dangerous areas. Hart’s shows his relative lack of mobility, with the xG per shot well into that 0.10-0.20 bucket from which far more goals are typically scored.

If you once again return to the initial radar, the CCAA% metric measured how often a keeper moved to attempt a claim relative to the average keeper. This showed Hart has been more aggressive than the average keeper in attempts, but his success rate of 81% lagged the league average of 85%. In comparison, Barkas was one of the most active keepers while also having been well above average in success rate. The ‘eye test’ seems to undervalue the importance of keepers being able to claim balls relative to shot-stopping. Former manager Lennon appears to have relied more on the eye test:

Celtic Way:

Celtic Way:

Bain’s heatmap for last season showed his lack of claiming actions, and his shot distribution against crosses was abysmal.

With the shot-stopping and claiming issues split between Hart and Barkas, what about passing and ball progression as the launching point for buildup and attacks?

Celtic Way:

Despite having significantly lower volume in passes and carries, those components of OBV were higher for Barkas last season - more with less. We can also see that he was one of the best in the league at passing under pressure, whereas Hart has been one of the worst. Despite playing in a significantly less potent attacking team, his xG Buildup was also higher.

Does all of this somehow suggest that Barkas or Hart are the ‘better keeper’? In a sense, we are comparing an apple to an orange given the disparity in playing styles and performance levels between the two Celtic iterations.

Ultimately, Celtic appear to have recruited Barkas to play like Forster, and then Hart to play like Barkas, with Bain now backing up Hart to play like Barkas instead of Barkas, only be worse at the role.

After a rough 18 months in Glasgow, Barkas appears, even if his shot-stopping remains mediocre, a good choice for a team looking for a sweeper keeper. Do you know of any who might need one?