It’s entirely reasonable to compare the total xGA of the games where these 9 goals were actually conceded. That provides a more accurate basis for analysis and comparison. Using the full 12-game sample size feels arbitrary and less relevant. Based on the actual comparison, Arsenal conceded a little under 2.5 more goals than their xGA suggests they should have. While this isn’t the same as the tweet—which I agree lacks context and nuance—it’s a more accurate approach than relying on an unrelated 12-game sample size. Don’t you think?
I am a fan of using as big a sample size as possible, so the more games the better. I’m not sure why Scott picked 12 games, but for me, 9 games is too few. I would look at xGA vs GA for the whole season to date.
Is this right Scott? Across the games these goals relate to, I make it 6.58 xGA yet we conceded 9 goals? Aren’t we on a very basic level 2.5 goals against worse off than we should be? Am I missing something here?
I get we are looking at a 12 game period, but that’s the wrong game tally for the goals mentioned? Shouldn’t we be comparing this against the actually xGA of the games in which we conceded the goals?
It is the same games. I am sure I am follow. The original point is the last 12 games in all comps, I included the information for the shots allowed for the last 12 games in all comps in this.
This is excellent analysis and a great explainer to people unfamiliar with statistics. Very well done.
I KNEW this was coming after seeing that silly xG per goal conceded “analysis” doing the rounds on Twitter the other day 😂
It’s entirely reasonable to compare the total xGA of the games where these 9 goals were actually conceded. That provides a more accurate basis for analysis and comparison. Using the full 12-game sample size feels arbitrary and less relevant. Based on the actual comparison, Arsenal conceded a little under 2.5 more goals than their xGA suggests they should have. While this isn’t the same as the tweet—which I agree lacks context and nuance—it’s a more accurate approach than relying on an unrelated 12-game sample size. Don’t you think?
I am a fan of using as big a sample size as possible, so the more games the better. I’m not sure why Scott picked 12 games, but for me, 9 games is too few. I would look at xGA vs GA for the whole season to date.
Is this right Scott? Across the games these goals relate to, I make it 6.58 xGA yet we conceded 9 goals? Aren’t we on a very basic level 2.5 goals against worse off than we should be? Am I missing something here?
I get we are looking at a 12 game period, but that’s the wrong game tally for the goals mentioned? Shouldn’t we be comparing this against the actually xGA of the games in which we conceded the goals?
I did use the xG in the last 12 games, from my model it was 8.5 xG from those matches
Ah okay got it, but isn’t the correct comparison the games in which those goals scored actually occurred?
It is the same games. I am sure I am follow. The original point is the last 12 games in all comps, I included the information for the shots allowed for the last 12 games in all comps in this.