SwansAnalytics
Vital Newbie
After developing my own expected goals model this weekend, I thought it would be worth breaking this down and explaining this ever-growing concept of "expected goals".
In simple terms, every shot on goal is given a rating between 0-1 on the probability of a shot resulting in a goal. For example, in the model I've developed, a penalty automatically gets a 0.75 rating because based on almost 44,000 shots since 2016/17, 3 of every 4 penalties are scored.
Those shots are from Premier League and Championship matches since 16/17 - rather than using older data or data from other Leagues. Using this data helps to have accurate averages for the level of football we're looking at here.
It's been criticised and misinterpreted over the years but it can provide some useful information on teams and players such as are they under-performing or over-performing? Looking at penalties again, 75% of penalties scored is the average, so anything over that is better than the average and vice versa.
It also helps to understand the quality of chances during a game. We had 31 shots against Stoke as an example, but how many of them were actually high quality shots?
How it works and how xG is calculated
There are a number of xG models out and one fair criticism of xG is that they will all output different calculated values.
Using a large number of shots data, we predict the outcome of a shot resulting in a goal based on:
Other models will involve more variables, defensive pressure, type of open play scenario (counter attack etc), assist type, cross type (high/low) etc.
Here is the xG values and shots map from yesterday's game at QPR:
Player xG
This might be of more interest. xG per player - in other words - who takes their chances and who is wasteful? Are players not converting chances they should be?
According to the model, McBurnie is scoring more goals than expected and that's not a great surprise over an entire season as he's scored some excellent and very low-scoring headed goals.
Dan James and Bersant Celina are 2 players that are under-performing. James, in particular, has missed some big chances this season but is beginning to make up for it now in recent weeks.
"It's Goals That Matter, Not xG!"
Of course, that's what wins games but what about all those chances that were missed/wasted? How can we rate chance creation? If Dan James puts the ball on a plate for mcBurnie and he missed, Dan James misses out and doesn't get an assist. That's where expected Assists comes in too but we'll leave that for another day!
In simple terms, every shot on goal is given a rating between 0-1 on the probability of a shot resulting in a goal. For example, in the model I've developed, a penalty automatically gets a 0.75 rating because based on almost 44,000 shots since 2016/17, 3 of every 4 penalties are scored.
Those shots are from Premier League and Championship matches since 16/17 - rather than using older data or data from other Leagues. Using this data helps to have accurate averages for the level of football we're looking at here.
It's been criticised and misinterpreted over the years but it can provide some useful information on teams and players such as are they under-performing or over-performing? Looking at penalties again, 75% of penalties scored is the average, so anything over that is better than the average and vice versa.
It also helps to understand the quality of chances during a game. We had 31 shots against Stoke as an example, but how many of them were actually high quality shots?
How it works and how xG is calculated
There are a number of xG models out and one fair criticism of xG is that they will all output different calculated values.
Using a large number of shots data, we predict the outcome of a shot resulting in a goal based on:
- Where the shot occurred
- Situation type (corner, open play, direct or indirect free-kick)
- body part (head or foot)
Other models will involve more variables, defensive pressure, type of open play scenario (counter attack etc), assist type, cross type (high/low) etc.
Here is the xG values and shots map from yesterday's game at QPR:
Player xG
This might be of more interest. xG per player - in other words - who takes their chances and who is wasteful? Are players not converting chances they should be?
According to the model, McBurnie is scoring more goals than expected and that's not a great surprise over an entire season as he's scored some excellent and very low-scoring headed goals.
Dan James and Bersant Celina are 2 players that are under-performing. James, in particular, has missed some big chances this season but is beginning to make up for it now in recent weeks.
"It's Goals That Matter, Not xG!"
Of course, that's what wins games but what about all those chances that were missed/wasted? How can we rate chance creation? If Dan James puts the ball on a plate for mcBurnie and he missed, Dan James misses out and doesn't get an assist. That's where expected Assists comes in too but we'll leave that for another day!
Last edited:
