Back in February, I introduced a non-shots expected goals model at the OptaPro Analytics Forum, drawing on earlier work from a year ago. I talked about a simple version of the ball progression model that’s now one of several evaluators used by NYA. As the name suggests, the ball progression model gives teams credit for moving the ball based on the chances of scoring goals from the resulting positions. How should we interpret its results?
The first generation of expected goals models attached a probability of scoring to each shot rather than to reaching specific positions on the field. These models were intuitive and useful; they captured performance in a way that was more persistent than counting actual goals. Yet there was a problem: What if a team was great at moving the ball into dangerous areas, but its strikers had a hard time getting shots off? The team’s expected goals in attack would look lousy, even though most of the players were doing their jobs well. This is why measuring ball progression is also worthwhile.
Today, both kinds of models can play an important part in analysis. As I’ve written before, we can use expected goals from shot creation and ball progression as a measure of efficiency. A team that generates 1.6 expected goals from ball progression but only 1.2 expected goals from shot creation may be suffering from the problem described above. Conversely, a team that creates 1.5 expected goals worth of shots from just 0.8 expected goals worth of ball movement may have an especially good cutting edge – perhaps an attacking midfielder who’s a through-ball genius. When this team moves the ball forward, it’s less likely that defenders will still be around when the ball arrives.
In either case, we’d want to use video to confirm the suspicions unearthed by the efficiency metric. Do the stories about strikers and attacking midfielders fit, or are we witnessing a phenomenon that’s not likely to be sustainable? The same is true on the defensive side. A highly efficient team – conceding more expected goals from ball progression than from shot creation – might be good at soaking up pressure and preventing shots; an inefficient one might have a back line that’s standing too far off the attackers when they’re trying to shoot.
If you’re reading this and thinking, “Wow, these could be really useful tools for talent identification as well as tactical analysis!” then you’re absolutely right. NYA uses versions of these models to rate players, too, in both attack and defense. But of course, two models can’t capture everything that’s important about a player or team. That’s why we also use other metrics – and why we see all of them as complements to traditional scouting and evaluation.