Celtic’s set-piece struggles and the role of Joe Hart

My friend, and fellow TCW columnist, Ross Goodwin recently offered an excellent Twitter thread in which he took an extensive look at both attacking and defending set-pieces.

This piece will hopefully add some additional perspective to his work, as well as take a more specific look at the goalkeeper position. All data and metrics will be only for league games.

Let us begin by reviewing some high-level performance metrics for set-pieces overall, which include indirect and direct free kicks, corners, and throw-ins (but not penalties):

We can see the trend of conceding higher amounts of goals from set-pieces via the darkest green bar, with just 0.11 goals per 90 minutes conceded in the 2018-2019 season. The next bar down represents xG conceded for set pieces, which for the 2018-2019 season was 0.18, with the third bar showing post-shot xG (PSxG). This is an important breakout, in my opinion, as it offers us the ability to decompose three vital elements:

1. Quality of chances conceded (xG)

2. Quality of finishing by opponent

3. Quality of keeper shot stopping

For example, in the 2018-2019 season, the 0.16 PSxG being below 0.18 xG shows that opponents’ poor finishing contributed 0.02, while shot-stopping 0.05 of the difference between goals conceded and xG conceded. 

If we replicate this exercise for each season, we can see the relative contribution for each of the three elements. The 2019-2020 season included the least amount of xG conceded, with last season by far the highest, and this season reverting back close to the level from 2018-2019. In each of the past three seasons, opponents have been finishing at a higher rate (PSxG higher than xG), while shot-stopping has been a negative contributor to varying degrees in all but the 2018-2019 season.

Here were the metrics for only corners, which offers additional nuance as to the relative performance levels each season:

As we focus upon what has transpired so far this season, a pretty clear picture emerges, in my opinion.

Obviously, conceding 0.33 goals per game from set-pieces is the worst of the five seasons for which we have these metrics, but the three elements offer additional clues as to potential issues.

Opponent finishing has contributed 0.05 and poor shot-stopping 0.09, with the latter standing out as almost double the 0.05 contribution of poor shot-stopping last season. 

Along with Joe Hart having issues with shot-stopping specifically on set pieces (he’s been good from open play), he has also been very conservative in patrolling his box.

Here were corners in which Celtic conceded shots for the 2019-2020 season, with Fraser Forster in goal, versus so far this season:

Some of the disparity could reasonably be attributed to the confluence of defenders, and not just keeper play.

For example, Christopher Jullien was aerially dominant in the 2019-2020 season, which likely played a factor. 

However, we can break out keeper claims specifically. Here was Forster for the 2019-2020 season:

StatsBomb breaks out the keeper box into nine different zones, and then tracks the number of claims per ball available to be claimed by the nine zones.

For example, zone 2.1 represents the central area between the penalty spot and the six-yard line, and for that specific zone, we can see that Forster averaged 0.019 claims per claimable versus a league average of 0.01.

This suggests he was more aggressive than average in ‘attacking’ balls into that area. In contrast, he did not venture into the three most distant zones (0.0, 4.0, and 1.2) at all. 

Now let us have a look at Hart so far this season:

This shows that Hart has similarly been conservative across distant zones, but it also shows how infrequent he has attacked balls in zone 2.1, with a rate around just 25 per cent of league average.

Here is another way to consider, which again shows how relatively anchored Hart has been centrally inside his keeper box:

This issue has begun to eat into the overall performance value Hart has offered so far this season:

Hart had been excelling with shot-stopping overall, but these issues with set pieces have now dropped that metric to the point where he’s prevented about the same number of goals (0.51) as would be expected given the shots he’s faced. Why is this a potential issue?

As the title race progresses, relative keeper play may become a match, and possibly even title-deciding.

Here is a comparison of Hart with Allan McGregor for goal prevention and distribution (note neither are particularly effective on the ball):

While McGregor remains prone to very poor positional errors, his shot-stopping has recovered from an early-season dip, while these issues on set pieces have dropped Hart to about league average in shot-stopping. 

Here are the two keepers by On-Ball Value (OBV):

We can see that McGregor’s issues with claiming and sweeping have almost wiped out the value of his shot-stopping, with his Goalkeeper OBV at 0.01.

However, Hart’s drop to around league average in shot-stopping have resulted in his Goalkeeper OBV dropping to -0.04.

To place that level into some perspective, Bain and Barkas’ Goalkeeper OBV for last season were -0.01 and -0.06. 

With much discussion about squad height and size, and it very plausibly playing some role in opponents’ PSxG being higher than xG for set pieces, Joe Hart’s issues with crossed balls appear to have been far more impactful.

Hopefully his shot-stopping from set-pieces improves and/or normal variance leads to opponents’ finishing to be poor over the duration of the season, as it appears crossed balls may remain an issue.