In the second half of their first knockout game, France unexpectedly found itself down 2-1 against Argentina. An upset was in the making, until French footballer Benjamin Pavard took this try from outside of the box.
According to a statistic from Opta Sports, the shot had an “expected goals” of just 0.03, meaning it would be expected to produce a goal three percent of the time. This figure is calculated from 10 different variables, but the two most significant for this particular shot are pretty straightforward. It was taken from 22 meters away from the goal line – pretty far out. And it was taken from 15 meters off the center line – a very bad angle.
To better illustrate the principle behind this sort of statistic, Opta provided us with 100 shots taken from a nearly identical location during Premier League, La Liga, Serie A and Ligue 1 play over the past 4.5 years. They mostly did not go in.
Drag to change the angle
While quantifying it can be nice, it is not exactly a shocking insight this was a difficult shot. The true usefulness of expected goals comes from aggregating a lot of shots together. Soccer is inherently a low-scoring game – while better teams generate more opportunities and prevent their opponents from doing the same, individual goals can be fluky. Expected goals can show us when a team is generally producing good shots and playing good defense, even if those efforts are not reflected in the final scoresheet.
Here is how that looks for all the World Cup teams in the tournament so far, updated through the semifinals.
The best expected goal differentials in the World Cup
|Expected goals||Actual goals|
Brazil, surprisingly, leads the pack by a wide margin with an expected goals differential of +9.3. Unfortunately, they were only able to score eight actual goals compared with an expected figure of 12.3, and failed to make the semifinals.
This gap can mean a lot of things. The team could have been unlucky, or they could have played poorly in such a way that they were unable to finish as many opportunities as you’d typically expect. (Or both!) On the flip side, surprise quarterfinalist Russia was the biggest overproducer, with an actual goal differential of +4 compared with an expected goal differential of -2.5.
What expected goals misses
As mentioned, two of the biggest components of the expected goals model are shot distance and angle. It also accounts for the “passage of play” (assisted, corner, free kick, etc.), whether the shot was off a rebound, and a couple other factors. You can see a complete list here.
The model has a couple weaknesses. For one, it does not account for individual players. A series of free kicks from the same spot would all get the same expected goal value, even if one of them was taken by Cristiano Ronaldo. And it does not account for the position of defenders at the time of the shot – outside of acknowledging 1-1 opportunities – although that may change in future iterations.
While it sounds obvious, expected goals only knows abouts shots. Beautiful passes, runs and touches can generate opportunities that do not quite produce a good attempt, but are nonetheless evidence a team is playing well. (Other models, such as the one used at FiveThirtyEight, try to account for these “non-shot” actions.)
The opposite is true too: If a team misses an easy shot, the failure will be reflected in its expected vs. actual goals. But if they blow what should have been an easy scoring opportunity thanks to a bad pass, expected goals never know about it. Take for example this situation that Mexico found itself in against Germany, which somehow did not end in a clean chance.
Nevertheless, expected goals is still a useful analytic tool, and like the best stats it often agrees with less quantified analyses of game play. England had a fluky group stage, narrowly beating Tunisia 2-1 thanks to a last second header, and then destroying Panama 6-1. Expected goals tells a more consistent story about the quality of the team: 3.2 expected goals to 1.1 in the match against Tunisia, 2.7 to 1.0 in the match against Panama.
About this story
Shot paths and expected goal data from Opta Sports.
Originally published July 5, 2018.