'Off the charts': Where Celtic are becoming a truly modern machine

While wage bills appear to be paramount, when clubs are within relatively close proximity to one another or resources are comparably flat as they are in the MLS and A-League via salary caps, the non-wage bill factors become determinant. 

These are:

– How well are the financial resources used (for instance, how good is the talent level of players recruited and developed)?

– What is the relative quality of coaching, in-game strategy, man management, selection and injury prevention/treatment, among other things?

– How does good, old-fashioned luck break?

As referenced previously, significant disparities in impactful officiating decisions such as red cards and penalties can be enough to materially impact the league table. 

In addition, teams do not control all factors. For example, so far this season, Celtic have enjoyed an unusually poor quality of play from opposing goalkeepers through the first 15 games. 

In fact, if this composite keeper were an actual player, he would be over 200 per cent worse than the actual worst keeper as measured by StatsBomb’s goalkeeper OBV metric. 

In contrast, the quality of keeper play Rangers have faced so far this season would rank second-best. These sorts of disparities tend to even out over the long term but can also have a meaningful impact on a 38-game season.

Outside of those ‘luck’ factors, how should we think about measuring and attributing between player-specific performance versus the impact of coaching decisions? Obviously some concepts – such as man management and leadership – are qualitative and not directly measurable. 

However, we can look at various performance metrics to try to piece together a better assessment and attribution. To continue the exercise from the last column, we will be comparing the 2018-19 and 2022-23 Celtic teams. 

From the prior column, we can see that Celtic’s staffing costs were approximately £56million for the 2018-19 season and, while this season's costs will not be reported publicly until sometime next autumn, for this exercise we will assume it will not be materially higher than last season. Assuming some typical industry inflation for player values and wages, the amounts are likely to be comparable. 

This first radar offers a simple view of the two teams across broad performance metrics. Obviously, we are only comparing a partial season to a full season, so this exercise is not ‘apples to apples.’ While not shown, the 2018-19 team’s metrics looked very similar to the season-long levels over the first 15 games. 

We can see from the radar that the current vintage of the Hoops has been ‘off the charts' better from an xG perspective, both created and conceded, as well as the average quality of chances for and against. 

Stylistically, Ange Postecoglou’s side has been more direct while playing a higher defensive line on average, although they have been less aggressive in counter-pressing. Those metrics and the description probably conform with how many supporters remember that 2018-19 team: a bit more methodical with more sideways passing and recycling of possession.

If the overall performance levels have been superior, and the financial resources have been comparable, then how can we consider player-specific value versus that of Postecoglou? 

This is just one relatively simple way to try to benchmark attacking contributions by players to team-level performance. Publicly available attacking performance data is far more advanced and developed, which is why we are focused on it in this exercise. 

We can see the various disparities and similarities by player and position groups, with strikers followed by wingers, midfielders and then full-backs. Due to the relatively low volume of attacking output, centre-backs are excluded from this exercise.

Keep in mind that the sample sizes for some of this season’s players are fairly small. This can be distortive for a player like Liel Abada, who has appeared in just about half of available league minutes, or Sead Haksabanovic with about a third.

Of course, shots and the passes which set up shots (and result in the expected assist metric) are not the be-all and end-all of measurable attacking performance. Ball progression and passes which precede a shot assist are obviously also of value. 

This graphic takes the attack-related OBV sub-components and shows them as a percentage of the overall team amount for each player. Some interesting outliers exist, with Leigh Griffiths and Emilio Izaguirre from the 2018-19 team notable. 

As with any analysis, it is important to consider potential explanations for outliers and, in these instances, they appear to have to do with a combination of sample sizes. In Griffiths’s case, extremely high-quality set-piece deliveries over the small sample. 

The remainder shows some evidence to suggest variations in style of play, as well as potential talent levels and breadth of skills. For example, we see evidence of Odsonne Edouard’s far more engaged style of play within that team’s system versus the limited time on the ball of strikers under Postecoglou. The flow of value through the wing position in the current team compared to going more through the midfield in the 2018-19 team also has some evidence. 

A robust modelling process using data over multiple seasons could assign per-player performance measurements using wages to generate ‘return on investment’ analysis. Benchmarking versus clubs and their players in other leagues playing similar styles, while also considering the wage bill structures of each league, could offer further context as to manager and player performance levels. 

READ MORE: ​How far does Celtic's wage bill guarantee success?

In addition, age-based development curves could be monitored in order to measure when player value may be maturing and incorporated into a player trading strategy. For example, 2018-19 was Edouard’s age-21 season, with his 2019-20 output growing to represent over 25 per cent of the xG + xA metric shown above, and 12.5 per cent of attacking OBV. 

Robust modelling and associated benchmarking could enable more efficient player acquisitions and sales. For example, much of Giorgos Giakoumakis’s production output appears to be driven by playing in Postecoglou’s system for Celtic versus domestic-level competition, whereas Jota’s idiosyncratic qualities appear evident even within this analytical framework. 

Attacking output is just one important aspect of football, with defensive and spatial aspects critically important as well, and a comprehensive analytical framework should obviously incorporate it.

As data quality continues to improve, the ability for clubs like Celtic to better measure and assess performance attribution could be key in enabling the club to compete in the Champions League and against sides with much greater resources.