In recent weeks I have started putting together my shot database to compile enough shots to work out xG. I’m putting a lot of time into it at the moment so posts have been few. I thought that I’d explain what expected goals is and my model.
Expected goals (xG) is a tool to value the probability of a shot using a regression of shots from a database. The database takes into account a variety of variables of the shot which will be covered later. The regression searches the database for a shot with the similar variables to get a probability of how likely the shot is to result in a goal.
ExpG quantifies the shot to how good it is. A team can have 30 shots in a game but they could all be from wide angles, near to the side lines with 11 defenders in front of him. While another team could take one shot from within the six yard box after winning the ball in the final third and playing a throughball and dribbling round the keeper. The second team is far more likely to win with these shots.
My model has 23 basic variables which I record for each shot I watch. Some have less impact but are useful to record for other research, e.g Game State. The variables:
- xy of Shot
- Type of Assist
- Counter Attack?
- xy of Goalkeeper
- Ball won in final 3rd?
- Minute of shot
- Score of Attacking team/Defending team
- Free Kick?
- Where placed in net
- Numbers in attack /defence
- Distance from goal
- Whether it results in a goal
I am using matches from many nations, leagues and levels of football. e.g. Gibraltar First Division, Champions League Qualifying rounds, World Cup Final, Premier League and National Conference League. It allows a fair representation of the probability of a goal, not just shots from world class strikers with world class playmakers assisting them.
You can add up the expected goals of each team’s shots from a game or a player’s shots over a game or season. It can show the ability of finishing from a team/player and can show luck which is a strong factor in football and life that is hard to quantify.
I hope you understand the theory of xG if you didn’t before. Please visit the great site 11tegen11 for more information and better analysis and examples about xG.
All feedback appreciated: sidelineteamtalk @ gmail [dot] com and follow me on twitter @SL_TeamTalk