Totally agree Simon. Love your information about English Football. In many ways it felt like a win in almost all sense except the scoreline. Which happens in Football (Soccer as many call it) that your side can dominate most stats except the scoreline. Made even more impressive when you think that Sicily and Newcombe were not strong contributors. Saw lots of positive signs and I expect a strong finish to the year from the Hawks. We will trouble every side this year and on days when we do make the most of our chances we will win more than we lose. Gave me even more sense that we are on the right track and just missing a few pieces of the jigsaw. Phillips must surely play this week to provide that extra tall in place of Finn. Using him as a sub was silly really. Would have made more sense to play him at the start tagging Merrett and then bring on a fast runner in the last quarter.
Thanks for a great article Simon. I believe that Champion Data does a metric called Expected Score and Simon Goodwin referred to it after their first final loss last year (the expected score indicated that Melbourne ought to have won). Both to the eye and borne out in a number of the key stats from Saturday's game, we were highly competitive, and it was a game that we could have won. Your xS analysis is just another stat that confirms that view and gives us further encouragement. What I liked most about the game, is that the things that didn't work, appear to be readily fixable, especially centre clearances which were a strength last season and set shot kicking for goal. While we improved from round one last year, whether we've really improved will be measured over a larger sample - here's hoping that we have a real crack against a seasoned finalist on Saturday.
"If you torture the stats, they will confess to anything."
No doubt analytics have a purpose, but in the end, they provide evidence/information to assess or analyse a performance rather than determine an outcome.
I don't know that you'd need a statistic to assess whether the team was competitive on Saturday.
I'd suggest the Amon miskick out of the back half, the lax marking of short leads in the defensive 50 & the failure to go with their opponent to the fall of the ball close to the goal square would be evidence that our attention to detail let us down.
As to the expected goal assessment, for games where there is no "reward" for missing, so it is a binary outcome - a goal or a miss - I see its direct relevance, but for the AFL, where a team can score a point for missing, there is some tweaking to be done. As the explainer acknowledges, the expected outcome is particularly relevant in low scoring games.
The other aspect in football (soccer) the goals are of a finite space (7.32m x 2.44m) whereas in AFL the goal posts are 6.4m apart (& based on the eye of an official as to whether the ball goes over the post or not), along with the height being infinite, I'd suggest make the assessment of expected goals a little harder to assess; like a lot of aspects of AFL.
Hey Mick. xS in footy obviously is/would not be perfect, just as it's not in soccer. But the fact that its scoring slides on a scale up to six 6 rather than 1 is not a drawback. Even if goals disproportionately weight the expected score upwards, it remains that the closer the xS gets to 6, the easier the shot is. I believe models could even account for things like type of shot (set shots/snap), levels of pressure, or right/left foot, but that data is not usually available to the public. And minor irregularities like field size and umpire error smooth out with enough data in the model. Of course, no model is perfect, and as you say, teams don't only generate shots from their own play (see Amon). xS would be best used as a long-term view of how a team has played relative to their opponent.
Last week, I had to raise an eyebrow when I heard an analyst apply this model to the Pies/GWS* game and suggest that in 7 out of 10 games if the same stats were applied that the Pies would win. I get the statistical application, but football, irrespective of the code, isn't an exact "science". Watching that game if the two sides played the exact same way, I'd think that the Swans would win at least 7 out of 10 times.
As the explainer article stated:
"Game-level xG
The main criticisms of expected goals often appear in scenarios where the metric isn’t being applied correctly. The most common of which is at the game level. A team having a higher xG total in a match doesn’t necessarily imply that they should’ve won the game. xG is only measuring chance quality and not the expected outcome of the game."
I take this to mean that this is the likelihood of kicking a goal as against missing it. For us, with Breust & Gunston missing shots that they are expected to kick, irrespective of the overall percentage across the league, really hurts the result of the game.
(I know that Fox often put up a graphic of individual players set shot goal kicking record from a spot on their broadcast as an indicator.)
This is where this calculation would align with the sport that could always have an expected score - cricket. Using the batting averages of the team and even more so in recent years where grounds, bowlers and innings of the game have been more easily incorporated into the calculation, but this is rarely applied in such specific detail as averages are just that - a measure of past performances, but not a predictor of today's performance.
The other aspect of the methodology outlined for soccer is the assessment of the role of the goalkeeper. As we know in AFL, there are effectively 18 goalkeepers coming from any direction on the ground, so I don't know how this is assessed for AFL. Similarly, the trajectory of a soccer ball could be assessed similar to the projection of a cricket ball in a DRS for LBW, however as we all know the bounce of a Sherrin is much more unpredictable as Saints' supporters well know.
Fundamentally, I see this concept being a consequence of the "money ball" effect where statistical evidence was deemed to be the "pure" form of analysis as against the "gut feel" of recruiters/scouts. However, "money ball" was based on assessing the performance of the individual rather than the team collective.
(A similar debate is playing out in the NFL with the debate of going for it on 4th down or taking a field goal attempt (FGA). If you fail on 4th down or miss the FGA, this becomes a turnover and field position whereas a successful FGA gains 3 points or a successful 4th down attempt gives another set of downs. This is a risk & reward analysis rather than accuracy.)
Where this analysis relates to shots at goal, it does makes sense I think. If a team has 6 shots at goal from a point on the ground where the AFL average is 50% and kicks 6 out on the full, that tells me that the team 'should' have kicked 18 points. While, as you say, it probably doesn't take account of who is taking the shot and that person's average performance for that particular shot, if the data sample on which it is based is large enough, the analysis does carry weight.
Agree with all this. I thought we were much more likely to win until the fourth quarter. Goal kicking aside, our defence couldn’t put enough pressure on to prevent Essendon making good opps. On another day if they’d kicked like we did we would have got away with it. Overall though it’s a big improvement YoY.
Whilst I like the notion of using Expected Score to predict what a team should have scored on average it doesn't take into consideration that the ball would go back to the middle and we were being smashed out of the middle which could lead to the opposition scoring even more.
A lot has been made of center clearance but to lose in the manner we did belies the fact that the Hawks could move the ball and enter Forward 50 in a manner that is new and better
The 2023 version would have lost this game by 10 goals or more. That is progress.
We are a 4 goal better side this year. Ginnivan and Chol give us something extra. Watson will improve.
As for the mid-field. It will come together. We clearly miss Will Day and Nash is not at his best. Would love to see HH given a run
I know HI loves Ned Reeves. He is not the answer. Hit outs to advantage is a pointless stat. The really good rucks give you something around the ground. We are effectively playing a man short. Very hard to win that way.
Centre clearances 18-9. Two-thirds of the time the game restarted deep in Essendon's forward line. That's why we lost and never really looked like winning to my eye.
Totally agree Simon. Love your information about English Football. In many ways it felt like a win in almost all sense except the scoreline. Which happens in Football (Soccer as many call it) that your side can dominate most stats except the scoreline. Made even more impressive when you think that Sicily and Newcombe were not strong contributors. Saw lots of positive signs and I expect a strong finish to the year from the Hawks. We will trouble every side this year and on days when we do make the most of our chances we will win more than we lose. Gave me even more sense that we are on the right track and just missing a few pieces of the jigsaw. Phillips must surely play this week to provide that extra tall in place of Finn. Using him as a sub was silly really. Would have made more sense to play him at the start tagging Merrett and then bring on a fast runner in the last quarter.
Thanks for a great article Simon. I believe that Champion Data does a metric called Expected Score and Simon Goodwin referred to it after their first final loss last year (the expected score indicated that Melbourne ought to have won). Both to the eye and borne out in a number of the key stats from Saturday's game, we were highly competitive, and it was a game that we could have won. Your xS analysis is just another stat that confirms that view and gives us further encouragement. What I liked most about the game, is that the things that didn't work, appear to be readily fixable, especially centre clearances which were a strength last season and set shot kicking for goal. While we improved from round one last year, whether we've really improved will be measured over a larger sample - here's hoping that we have a real crack against a seasoned finalist on Saturday.
"If you torture the stats, they will confess to anything."
No doubt analytics have a purpose, but in the end, they provide evidence/information to assess or analyse a performance rather than determine an outcome.
I don't know that you'd need a statistic to assess whether the team was competitive on Saturday.
I'd suggest the Amon miskick out of the back half, the lax marking of short leads in the defensive 50 & the failure to go with their opponent to the fall of the ball close to the goal square would be evidence that our attention to detail let us down.
As to the expected goal assessment, for games where there is no "reward" for missing, so it is a binary outcome - a goal or a miss - I see its direct relevance, but for the AFL, where a team can score a point for missing, there is some tweaking to be done. As the explainer acknowledges, the expected outcome is particularly relevant in low scoring games.
The other aspect in football (soccer) the goals are of a finite space (7.32m x 2.44m) whereas in AFL the goal posts are 6.4m apart (& based on the eye of an official as to whether the ball goes over the post or not), along with the height being infinite, I'd suggest make the assessment of expected goals a little harder to assess; like a lot of aspects of AFL.
Hey Mick. xS in footy obviously is/would not be perfect, just as it's not in soccer. But the fact that its scoring slides on a scale up to six 6 rather than 1 is not a drawback. Even if goals disproportionately weight the expected score upwards, it remains that the closer the xS gets to 6, the easier the shot is. I believe models could even account for things like type of shot (set shots/snap), levels of pressure, or right/left foot, but that data is not usually available to the public. And minor irregularities like field size and umpire error smooth out with enough data in the model. Of course, no model is perfect, and as you say, teams don't only generate shots from their own play (see Amon). xS would be best used as a long-term view of how a team has played relative to their opponent.
Hey Simon
It's a good discussion point.
Last week, I had to raise an eyebrow when I heard an analyst apply this model to the Pies/GWS* game and suggest that in 7 out of 10 games if the same stats were applied that the Pies would win. I get the statistical application, but football, irrespective of the code, isn't an exact "science". Watching that game if the two sides played the exact same way, I'd think that the Swans would win at least 7 out of 10 times.
As the explainer article stated:
"Game-level xG
The main criticisms of expected goals often appear in scenarios where the metric isn’t being applied correctly. The most common of which is at the game level. A team having a higher xG total in a match doesn’t necessarily imply that they should’ve won the game. xG is only measuring chance quality and not the expected outcome of the game."
I take this to mean that this is the likelihood of kicking a goal as against missing it. For us, with Breust & Gunston missing shots that they are expected to kick, irrespective of the overall percentage across the league, really hurts the result of the game.
(I know that Fox often put up a graphic of individual players set shot goal kicking record from a spot on their broadcast as an indicator.)
This is where this calculation would align with the sport that could always have an expected score - cricket. Using the batting averages of the team and even more so in recent years where grounds, bowlers and innings of the game have been more easily incorporated into the calculation, but this is rarely applied in such specific detail as averages are just that - a measure of past performances, but not a predictor of today's performance.
The other aspect of the methodology outlined for soccer is the assessment of the role of the goalkeeper. As we know in AFL, there are effectively 18 goalkeepers coming from any direction on the ground, so I don't know how this is assessed for AFL. Similarly, the trajectory of a soccer ball could be assessed similar to the projection of a cricket ball in a DRS for LBW, however as we all know the bounce of a Sherrin is much more unpredictable as Saints' supporters well know.
Fundamentally, I see this concept being a consequence of the "money ball" effect where statistical evidence was deemed to be the "pure" form of analysis as against the "gut feel" of recruiters/scouts. However, "money ball" was based on assessing the performance of the individual rather than the team collective.
(A similar debate is playing out in the NFL with the debate of going for it on 4th down or taking a field goal attempt (FGA). If you fail on 4th down or miss the FGA, this becomes a turnover and field position whereas a successful FGA gains 3 points or a successful 4th down attempt gives another set of downs. This is a risk & reward analysis rather than accuracy.)
Where this analysis relates to shots at goal, it does makes sense I think. If a team has 6 shots at goal from a point on the ground where the AFL average is 50% and kicks 6 out on the full, that tells me that the team 'should' have kicked 18 points. While, as you say, it probably doesn't take account of who is taking the shot and that person's average performance for that particular shot, if the data sample on which it is based is large enough, the analysis does carry weight.
Agree with all this. I thought we were much more likely to win until the fourth quarter. Goal kicking aside, our defence couldn’t put enough pressure on to prevent Essendon making good opps. On another day if they’d kicked like we did we would have got away with it. Overall though it’s a big improvement YoY.
Whilst I like the notion of using Expected Score to predict what a team should have scored on average it doesn't take into consideration that the ball would go back to the middle and we were being smashed out of the middle which could lead to the opposition scoring even more.
Agree
We are going to see a bit of this in 2024
Not everything will work all at once
It is part of the development process
A lot has been made of center clearance but to lose in the manner we did belies the fact that the Hawks could move the ball and enter Forward 50 in a manner that is new and better
The 2023 version would have lost this game by 10 goals or more. That is progress.
We are a 4 goal better side this year. Ginnivan and Chol give us something extra. Watson will improve.
As for the mid-field. It will come together. We clearly miss Will Day and Nash is not at his best. Would love to see HH given a run
I know HI loves Ned Reeves. He is not the answer. Hit outs to advantage is a pointless stat. The really good rucks give you something around the ground. We are effectively playing a man short. Very hard to win that way.
Centre clearances 18-9. Two-thirds of the time the game restarted deep in Essendon's forward line. That's why we lost and never really looked like winning to my eye.
We didn't. Nothing changes that. Fight like hell next week.