As Sean Martin introduced in one of his columns last week, Statsbomb released a new metric titled “on-ball value” (OBV). I encourage readers to review Sean’s summary and follow the link to Statsbomb’s explanation of the metric for additional information. In my analytical experience, this is a natural progression following the path of other sports, with baseball the pioneer ala Moneyball.
Baseball is a far easier game to quantify due to its discrete actions and events, so valuing the contribution of each action relative to runs scored is relatively straightforward. However, it has been an evolution as technology and related modeling have advanced. For example, a metric titled “Wins Above Replacement” (WAR) became ubiquitous a number of years ago as a sort of all-in-one metric for each players’ contribution to performance – i.e., value.
Advances in camera and tracking technology introduced another generation of data and related metrics. For example, defence-related metrics remained very much disputed, even within the context of WAR, but now more advanced data has greatly improved the ability for players’ defensive value to be measured.
Much of this echoes with the evolution in football data and related analytics, though football’s relatively late start has resulted in more accelerated changes. American Soccer Analysis introduced Matthias Kullowatz’s goals added (g+) model in May 2020, and my understanding is that Statsbomb’s OBV is in the same neighbourhood. Other similar models are listed in the Statsbomb link provided above. I consider these in aggregate to be comparable to the evolution to WAR in baseball, which is to say a very important leap forward and improvement but still missing extremely important aspects. Specifically in football, and example would be off-the-ball and positional movements.
These are important concepts due to the enhanced ability to better value components of the game such as ball progression, risk-benefit analysis of passing, and also the impact of things like defensive lapses and where they occur.
Michael Stewart’s recent column expressing his views that Celtic’s current first-choice center-back pairing are not good enough at the European level created a decent amount of discussion amongst our support, including amongst us stat nerds. I thought it was a good opportunity to get more familiar with OBV, and also examine whether there is much performance evidence to either support or contradict Stewart’s argument.
Prior to OBV, I typically broke performance radars down into phases of play such as attack, defence/press, and passing. Here are combined radars for Carter-Vickers and Starfelt for league games so far, showing my custom attack, defending, and passing versions:
One could understandably look at the pervasiveness of high-percentile rankings and consider both to have been relatively dominant. These radars are focused upon what I call volume and efficiency metrics. For example, how many aerial duels and what % of them are “won.” For related metrics in the defensive radar, we can see that Starfelt has had some issues, whereas Carter-Vickers has been very efficient.
The disparity in their xG per shot under their attack radar also suggests superiority by Carter-Vickers relative to getting on the end of corners and set pieces. Looking for patterns and relationships via an ensemble approach is where the analytics domain enters after quality data is obtained.
Next let’s take a look at comparable radars for Europa League games, but first an important analytical judgment call to explain. I’ve limited Starfelt’s sample to exclude the Jablonec games, as they are a level or two down in quality of opponent relative to the remainder of the sample. Despite being in the playoff round, I did include AZ Alkmaar for Starfelt because of their relative competitive quality.
Here we see some degradation within impact on attack, which is to reasonably be expected given the disparity in domestic dominance versus this level. Passing metrics remain very high in volume and still hold up pretty well in efficiency, and again show a drop in attacking involvement. Defence metrics show volume and efficiencies that are generally good to okay at this level, with Starfelt’s issues in the air having persisted.
On-ball value (OBV) enters the picture to try and help answer the following question: How do we balance all this out?
What is the relative value of all of the individual events each player has engaged in relative to either improving the probabilities of Celtic scoring goals, but also preventing the opposition from doing the same? Here is my custom OBV-related radar for domestic games and then the Europa League sample already used:
Well, now I am confused! Obviously, we are dealing with relatively small sample sizes here, so that is a very important consideration. Given that the Defensive Action OBV metric weights things like the importance of errors, Starfelt’s various early season blunders appear to have been picked up in his domestic version. In contrast, Carter-Vickers has had a rougher time of it defensively in the Europa League. At first look, they both seem to be okay on Passing OBV and Dribbles & Carries OBV, while the number of shots for them in such a small sample make that metric lacking. So overall, does this mean that Stewart was “wrong?”
This is where the analytics domain and looking for logical inferences and patterns may help. I added reasonable proxies for OBV-related volume metrics within the radar. For example, for the Passing OBV, I added Open Play Passes as well. For what purpose? Back to a baseball we go!
Voting, selection, and entry into the Baseball Hall of Fame is a big deal amongst some of us older baseball fans. Historically, it has been what some describe as a “small room” Hall of Fame, which is to say sports like the NFL and NBA have been far more expansive with selection criteria. Baseball Hall of Fame voters, which have historically been newspaper writers covering the sport, have been pretty stingy. This has led to all sorts of debates over the years, with a descriptor emerging – “the compiler.”
A compiler is a player who accumulates a lot of counting stats due to longevity rather than any specific greatness in individual seasons, or in an overall career. For example, there have been some players who have played 20+ seasons and been good, if not great players, who have compiled huge counting statistics. Some breached rarified air like 3,000 hits, even having not been great hitters.
I mention all of this because my assessment of both Carter-Vickers and Starfelt and their OBV-related metrics strikes me as having elements of “compilers.” They happen to play in a system that results in a high volume of passes, which is reflected in their high percentile ranking even in Europa League. However, compare those volume-focused percentiles with their Passing OBV. A similar pattern emerges across the Defensive Action and Dribbles & Carries OBV metrics – their volume percentiles are a good bit higher than the OBV percentiles.
This is where a totality of an analytical approach can also offer valuable context. I had done a fair amount of analysis of both players when they were signed and looked at their performance histories at prior clubs. Neither had a profile that indicated a strong likelihood of being what many would consider “ball playing center-halves,” let alone at a European level.
This leads me back to my Jenga Tower analogy. As we witnessed again versus Aberdeen this past Sunday, the deficiencies of our keeper and centre-backs relative to buildup play can create issues relative to breaking opponents’ pressing. This has been amplified against European level opponents. Despite 95th+ percentile ranking for all three in both number of open play passes and carries, all are in low single digits in related OBV metrics.
In fact, despite it being heretical to some in the modern game, a strong case can likely be made to go long on goal kicks, but that is for another day.
As for today’s topic at hand, I agree with Stewart’s conclusion, but think his rationale is wrong. Celtic are ranked dead last in the Europa League group stage so far in conceding shots off opponents’ high press and 24th for shots off of counter-attacks. Building up from the back versus competent and athletic pressing sides, as well as having the athleticism to play a high line and deal with counter-attacks, are vital to the style of play Ange Postecoglou wants to play.
In my view, neither Carter-Vickers nor Starfelt fit the bill.
Comments: Our rules
We want our comments to be a lively and valuable part of our community - a place where readers can debate and engage with the most important local issues. The ability to comment on our stories is a privilege, not a right, however, and that privilege may be withdrawn if it is abused or misused.
Please report any comments that break our rules.
Read the rules here