xG statistics in football

What is xG

The term xG means "expected goals. Simply put, it is an indicator that calculates the probability that a goal kick will result in a goal. By referring to the number of shots taken from different positions (and their danger) and the information about which ones resulted in a goal, the expected goals model shows us how many goals a team can score on average.

Here is an example of pari match bookmakers. The value of xG equal to 1.00 indicates a guaranteed goal, that is with 100% probability, but in practice it is impossible. The chance of scoring a goal with xG being 0.40 tells us that if that shot had been taken 10 times, the ball would have hit the net 4 times. Although in some cases other numbers are taken instead of shots on goal, this is the main figure for determining how strong a team is.

Techniques for predicting expected goals

Each shot is assigned a certain danger level. For example, the closer the shot is to the goal, the higher the xG, as would be expected. Meanwhile, the expected goals rate is lower when using the head than when using the foot. To calculate xG, the following indicators apply:

  • type of assist (if a player is rolled to an empty goal or taken out 1 on 1 with the goalkeeper, of course, it is much easier for him to score than if he had to circle and only then shoot);
  • the angle of the shot (the sharper it is, the harder it is, the lower the xG)
  • the distance to the goal;
  • whether the head was used;
  • whether pressure was exerted by the defenders;
  • whether it was a standard;
  • the form of the player;
  • the skill of the goalkeeper;
  • the condition of the turf, and more.

Effective xG models

There are these basic models for calculating xG:

  • Michael Caley Model. One of the two (along with tegen11) models whose parameters can be found in the public domain. It takes into account how often a specific player kicks a goal, where the attack starts, if there was a circumvention before the kick, the distance travelled with the ball and much more.
  • The tegen11 model.
  • The Statsbomb model takes into account more defensive actions than others.
  • Understat model. Here, too, each of the moments is assigned a degree of danger from 0 to 1, then all the scores are added together. But the hazard criteria are slightly different than in the three models mentioned above.