Championship Expected Goals Table

[cheshire_red]

This is @InfogolApp’s Championship table based on expected goals. For those who aren’t too familiar with XG... If your club’s near the bottom, it usually means they conceded a lot of chances and didn’t create many either. If your side is near the top, then it’s the opposite!

 

[IAmNick]

Interesting. I saw this one from Experimental 361 which has us at the bottom:

Not sure of the difference? I thought XG was more about how many you'd be expected to score rather than concede - maybe your version has factored conceding in as well to create their table?

Also looking forward to a few people with 3 individual examples of XG being completely wrong turning up and declaring the entire idea of statistics in general to be pointless!

[Davefevs]

And here’s the one from Wyscout:
 

(Suggests that a lot of draws are given out)

[mozo]

Shameful!

Note that Wigan and Preston are a lot higher than us in both

[Davefevs]

Noted.....first thing I looked at!

[The Bard]

This paints a worrying picture.  It says we have managed, somehow, to achieve better results than our underlying performance suggests.  

Why have we done better than these particular stats suggest we should? Did we have some underlying doggedness that saw us through even though we weren't actually that good?  Is that a particular Johnson characteristic because that is exactly what the 2007/08 season felt like too. 

One wonders how much attention the likes of JL and MA paid to this.  

 

[RoystonFoote'snephew]

So basically none of these tables bore any relation to our finishing position. 

[mozo]

9 minutes ago, The Bard said:

This paints a worrying picture.  It says we have managed, somehow, to achieve better results than our underlying performance suggests.  

Why have we done better than these particular stats suggest we should? Did we have some underlying doggedness that saw us through even though we weren't actually that good?  Is that a particular Johnson characteristic because that is exactly what the 2007/08 season felt like too. 

One wonders how much attention the likes of JL and MA paid to this.  

 

It appeared that way when we watched the games!

[MarcusX]

11 minutes ago, mozo said:

It appeared that way when we watched the games!

Don’t forget we had a fantastic start, if our first 10-15 results had been in the same vein as the rest of the season I think we’d have been close to the relegation battle

[ExiledAjax]

24 minutes ago, The Bard said:

This paints a worrying picture.  It says we have managed, somehow, to achieve better results than our underlying performance suggests.  

Why have we done better than these particular stats suggest we should? Did we have some underlying doggedness that saw us through even though we weren't actually that good?  Is that a particular Johnson characteristic because that is exactly what the 2007/08 season felt like too. 

One wonders how much attention the likes of JL and MA paid to this.  

 

Basically. Luck.

I'm a sucker for an xG thread. I feel like I have typed this response a dozen times on this forum, but it's worth doing.

xG seeks to help football fans/analysts quantify the affect that luck has on a club's position in the table. It is an attempt to test two old sayings firstly that "the table never lies". The table does lie. It happens fairly often. Secondly that "it all evens out across a season". It doesn't.

Basically football is so low scoring, and consists of so many interweaving variables, that an ounce of luck can go a very long way. Equally a bit of bad luck can hit a team very hard in the points column. Fans almost universally downplay the role of luck. In fact, humans downplay the role of luck. Luck can be very obvious: goals can be scored from deflections off of beach balls, bad referee decisions can lead to penalties, red cards can swing games from one result to the other. But generally luck in football is subtler: the bobble of the ball, a gust of wind, a striker just not quite hitting the ball as sweetly as normal. That's the kind of luck xG seeks to identify. The low scoring nature of the game, and the huge effect of luck is, for many fans, crucial to the enjoyment of the game, but it can also mean that people sometimes don't see how truly good/bad a team is - because it's easy to be blinded by the results.

xG does not seek to predict games, nor does it profess to state what a team 'should' have scored, or the results a team 'should' have achieved. What it hopes to do is stop clubs hiring and firing managers based purely on results. Some great examples are Newcastle giving Pardew an 8 year contract based on a 5th place finish in 2012, a finish which an xG analysis would have said was ludicrously lucky, or Dortmund firing Klopp in 2015 when Dortmund were struggling. xG suggested he and his squad had been massively unlucky in that first part of the season.

Closer to home you can go and look at the first ten games of this season just gone, In those 12 we scored 17, conceded 13, and chalked up 17 points. We were 7th and everyone was all aboard the play-off bus. However, xG suggested that generally a team playing the way we did and having and allowing the shots we did would reasonably expect to have scored 13, conceded 15, and gained only 12 points. We'd have been 15th.

Alternatively look at the other two notable streaks that LJ had with us, namely the record losing run and the run at the start of 2019 when we won 9 league games on the bounce. On each of those occasions the xG suggested that we had gotten either unlucky or lucky. xG supported keeping him during the losing run and supported the idea of not getting carried away during the two positive streaks.

There are other, sometimes simpler metrics that you can consider if you don't like xG. Foremost and simplest is goal difference. If a team finishes a season in position where its GD look out of place...then it's reasonable to assume they got there through some lucky, gritty, low scoring games. Alternatively they may have lost games like that. In many ways xG is simply a refinement of GD.

For context LJ's eventual sacking was pretty heavily supported by the xG for 2020.

3 minutes ago, MarcusX said:

Don’t forget we had a fantastic start, if our first 10-15 results had been in the same vein as the rest of the season I think we’d have been close to the relegation battle

The point is the performances were in the same vein as the rest of the season, we just got some luck. You're right that had results followed performances for the whole season we'd likely have been down in the bottom half dozen positions. As I said above, this isn't luck as in "I can't believe we got that penalty". It's luck as in "shots like that generally don't go in, but ours did this time" or "normally our opponent would score that, but on this occasion he's shot wide."

[ExiledAjax]

56 minutes ago, IAmNick said:

Interesting. I saw this one from Experimental 361 which has us at the bottom:

Not sure of the difference? I thought XG was more about how many you'd be expected to score rather than concede - maybe your version has factored conceding in as well to create their table?

Also looking forward to a few people with 3 individual examples of XG being completely wrong turning up and declaring the entire idea of statistics in general to be pointless!

Difference comes from the precision of the xG number, and the number of variables taken into account. I look at Infogol and E361 and average the two (hoping to add WyScout for next season) to get a reasoned view of our xG.

[MarcusX]

Just now, ExiledAjax said:

Basically. Luck.

I'm a sucker for an xG thread. I feel like I have typed this response a dozen times on this forum, but it's worth doing.

xG seeks to help football fans/analysts quantify the affect that luck has on a club's position in the table. It is an attempt to test two old sayings firstly that "the table never lies". The table does lie. It happens fairly often. Secondly that "it all evens out across a season". It doesn't.

Basically football is so low scoring, and consists of so many interweaving variables, that an ounce of luck can go a very long way. Equally a bit of bad luck can hit a team very hard in the points column. Fans almost universally downplay the role of luck. In fact, humans downplay the role of luck. Luck can be very obvious: goals can be scored from deflections off of beach balls, bad referee decisions can lead to penalties, red cards can swing games from one result to the other. But generally luck in football is subtler: the bobble of the ball, a gust of wind, a striker just not quite hitting the ball as sweetly as normal. That's the kind of luck xG seeks to identify. The low scoring nature of the game, and the huge effect of luck is, for many fans, crucial to the enjoyment of the game, but it can also mean that people sometimes don't see how truly good/bad a team is - because it's easy to be blinded by the results.

xG does not seek to predict games, nor does it profess to state what a team 'should' have scored, or the results a team 'should' have achieved. What it hopes to do is stop clubs hiring and firing managers based purely on results. Some great examples are Newcastle giving Pardew an 8 year contract based on a 5th place finish in 2012, a finish which an xG analysis would have said was ludicrously lucky, or Dortmund firing Klopp in 2015 when Dortmund were struggling. xG suggested he and his squad had been massively unlucky in that first part of the season.

Closer to home you can go and look at the first ten games of this season just gone, In those 12 we scored 17, conceded 13, and chalked up 17 points. We were 7th and everyone was all aboard the play-off bus. However, xG suggested that generally a team playing the way we did and having and allowing the shots we did would reasonably expect to have scored 13, conceded 15, and gained only 12 points. We'd have been 15th.

Alternatively look at the other two notable streaks that LJ had with us, namely the record losing run and the run at the start of 2019 when we won 9 league games on the bounce. On each of those occasions the xG suggested that we had gotten either unlucky or lucky. xG supported keeping him during the losing run and supported the idea of not getting carried away during the two positive streaks.

There are other, sometimes simpler metrics that you can consider if you don't like xG. Foremost and simplest is goal difference. If a team finishes a season in position where its GD look out of place...then it's reasonable to assume they got there through some lucky, gritty, low scoring games. Alternatively they may have lost games like that. In many ways xG is simply a refinement of GD.

For context LJ's eventual sacking was pretty heavily supported by the xG for 2020.

The point is the performances were in the same vein as the rest of the season, we just got some luck. You're right that had results followed performances for the whole season we'd likely have been down in the bottom half dozen positions. As I said above, this isn't luck as in "I can't believe we got that penalty". It's luck as in "shots like that generally don't go in, but ours did this time" or "normally our opponent would score that, but on this occasion he's shot wide."

I thought we played alright first few games but results may have clouded my judgement!

that said I remember watching the Reading home game and being absolutely bored to tears despite winning. I don’t know how we didn’t concede at the end, I was texting my mates telling them to bet on Reading to score so maybe you’re right about the luck aspect

[ExiledAjax]

Just now, MarcusX said:

I thought we played alright first few games but results may have clouded my judgement!

that said I remember watching the Reading home game and being absolutely bored to tears despite winning. I don’t know how we didn’t concede at the end, I was texting my mates telling them to bet on Reading to score so maybe you’re right about the luck aspect

Yeh we had some better performances. In my opinion xG is at it's best when you're looking at a larger number of games (ideally 30+), where you are looking for trends, or patterns. I wouldn't even start to look at it until about 10 games into a season. Every team can have a good or bad, lucky or unlucky, 90 minutes and really analysing a single game is better done through specific video and stats. 

 

[ExiledAjax]

13 minutes ago, JonDolman said:

Surely we were fantastic in quite a number of those early games.

I can't see how it is luck that won us those games. We looked a different team to after we lost Afobe to injury.

Possibly. I'd suggest that "fantastic" is a slight deviation into the world of hyperbole. We certainly weren't abject though.

But this is the whole point of xG: to test whether what you perceived with your eyes, and what the table told you - that we were, at times, fantastic - is borne out in the hard light of statistical analysis. 

Generally it wasn't.

Afobe got injured after the Boro game right? Up to that point we were on a cumulative xG of -0.82, averaging -0.14 per game. After that game we averaged -0.35 until Afobe returned against Blackburn, thereafter we averaged -0.1. So yes by the looks of it we were better when Afobe was available (he didn't play every game post-lockdown, and we had Wells in those matches)...but we were a long way from fantastic. 

Honestly when it comes to Afobe we are dealing with such a small sample as to make the stats almost meaningless...but would Afobe have maintained 1 goal every 2 games across the season and fired us into the top 6...I wouldn't have (and didn't) bet on it.

[Lrrr]

1 hour ago, IAmNick said:

Interesting. I saw this one from Experimental 361 which has us at the bottom:

Not sure of the difference? I thought XG was more about how many you'd be expected to score rather than concede - maybe your version has factored conceding in as well to create their table?

Also looking forward to a few people with 3 individual examples of XG being completely wrong turning up and declaring the entire idea of statistics in general to be pointless!

This is basically telling us we scored 5.3 more goals than we should have, so you've got clinical finishing in places we shouldn't score, Brownhill v Cardiff and Diedhiou's header would both have very low xG scores. We haven't taken many shots all season but tried working it into places where we'd have a good chance of scoring so the lack of shots means xG doesn't add up with them. On the defence we've conceded 7.6 less than we should have this could be bad finishing from the opposition but we've also seen Bentley pull of some ridiculous saves in all reality he shouldn't have.

[Pickle Rick]

1 hour ago, RoystonFoote'snephew said:

So basically none of these tables bore any relation to our finishing position. 

They kind of do actually.

We were overachieving being top 6 based on our stats. I'd say we were seriously riding our luck.

It eventually subsided to the poor form and drop closer to what we were 'achieving' in games.

Basic term, the team were not performing well enough and it near enough balanced out. There was no way we were pushing it as a better team in the league, to say the least.

I expect Jens Hegeler saw it coming. He would've forseen the likelihood of finishing even in the top 10 on match frequencies. Wouldn't surprised me if LJ had even been in communication with him.

[Lrrr]

11 minutes ago, Pickle Rick said:

I expect Jens Hegeler saw it coming. He would've forseen the likelihood of finishing even in the top 10 on match frequencies. Wouldn't surprised me if LJ had even been in communication with him.

Hegeler created the packing metric because he and another player felt they didn't get enough credit from normal statistical outputs so created something they thought would make them look better. 

[Hxj]

All these statistical analysis tools suffer from two major failings.

Firstly the prejudices of those who invent them. Nearly every goal is scored from a different place in different circumstances and with different teams.  How do you group these to produce a statistic?

Secondly the interpretation of the results, is appearing significantly below your league position good or bad? Is it as a result of clinical finishing, a fluke or a model that simply doesn’t work?

 

[Lrrr]

FBref explanation

https://fbref.com/en/expected-goals-model-explained/

[ExiledAjax]

1 hour ago, JonDolman said:

Does XG consider the shape of the defensive team? Where the defensive players are positioned when the attacking player has the shot?

Yes some models do. 

I've pondered it and I think perhaps the word "luck" conjures the wrong idea. Maybe "fortune" or "probability" better describe what xG seeks to quantify.

56 minutes ago, Hxj said:

All these statistical analysis tools suffer from two major failings.

Firstly the prejudices of those who invent them. Nearly every goal is scored from a different place in different circumstances and with different teams.  How do you group these to produce a statistic?

Secondly the interpretation of the results, is appearing significantly below your league position good or bad? Is it as a result of clinical finishing, a fluke or a model that simply doesn’t work?

Basically a grid system tracks where. The figures draw on a number of data sets, including historic analysis of tens of thousands of shots by thousands of players.

Every season this kind of data gets refined and shaped into a more sophisticated system.

Finishing in a different place in an xG model is neither good nor bad. What it can be is indicative of something going on behind the scenes or "under the hood" that suggests the position you find yourself in is neither sustainable nor repeatable (given the same parameters).

[Hxj]

53 minutes ago, ExiledAjax said:

Finishing in a different place in an xG model is neither good nor bad. What it can be is indicative of something going on behind the scenes or "under the hood" that suggests the position you find yourself in is neither sustainable nor repeatable (given the same parameters).

Or another possibility is that as the modelling tends to the average it predicts the answer for teams that play in the average way best.  As with all statistical models applied to the real world there will always be cases where the modelling simply doesn’t work, or predicts failure consistently despite the on field performances being the only result that matters.

[Mr Popodopolous]

Wigan's XG numbers being midtableish- dunno about 4th- but midtableish certainly seems about right! Doesn't factor in admin out of the blue with 3 weeks of the season left obviously and nor can it...

Always had a feeling too that Barnsley seemed better performance wise than results. We were lucky IMO to take 4 points off them for a start!

Leeds top- don't think we can argue tbh. Interesting that all 3 had Brentford 2nd and performance level possibly should've been but they seemed to blow up v Stoke and Barnsley! Had they been a bit more cautious and gone for the point at Stoke, they may well have beaten Barnsley in any case but the top 3 certainly right.

As for us. yeah below 12th seems right. Don't think I agree with the edge of the drop position or bottom 3 position but some of our underlying numbers were terrible! For good chunks of the season, results were outstripping performances and when that happens it often feels unsustainable.

Felt like even first 10,11, first 4 months or so our performances were not in keeping with our results, even when things were going well I thought.

[pnefcok]

For those wondering about Alex Neil's capabilities you only have to look at some of these xG/data table that are good performance indicators.

We would easily have been in the play-offs this season if we had any half decent striker at the club. Same in his first season in 17/18 when we sold Hugill on deadline day in January. We finished a point off 6th.

[ExiledAjax]

39 minutes ago, Hxj said:

Or another possibility is that as the modelling tends to the average it predicts the answer for teams that play in the average way best.  As with all statistical models applied to the real world there will always be cases where the modelling simply doesn’t work, or predicts failure consistently despite the on field performances being the only result that matters.

Completely fair criticism. I'd never profess that xG should be used in isolation to analyse football. That's totally incorrect. Every season there are anomalies that buck the model. Our club is one to bring that this year.

All I know is that I've followed this stuff for the best part of 5 seasons as it's developed and been adopted by many clubs, journalists and fans. Over that time it has generally proven itself useful to me in keeping my feet grounded and my head level as the rollercoaster that is Bristol City rolls on.

Sorry, I meant to just quickly explain xG to a forum member who asked what it was, I've ended up dominating the thread.

[Davefevs]

4 hours ago, JonDolman said:

Surely we were fantastic in quite a number of those early games.

I can't see how it is luck that won us those games. We looked a different team to after we lost Afobe to injury.

Not fantastic, but not bad either.  That back 3 (5212) system early season was able to absorb not having the ball, and push the opposition away from the penalty area, but being narrow and condensed.  Both wingbacks were happy to stay narrow (narrow for wing-backs) and allowed wide crosses.  They rarely allowed crosses from narrow positions.  As the CBs and CMs were also narrow (Palmer dropping in), the penalty area tended to be congested....and we got good blocks in on shots from around the 18 yd box.  Derby (a) was a good example.  I actually thought we played well with the ball that night, but we were structured without it.

Wyscout data from that game:

 

Stats Zone shot map for Derby:

For me that game was a classic “don’t just look at the stats, watch the game, then look at the stats”!!  It took a blinder of a goal late on from Marriott to give them hope.

4 hours ago, ExiledAjax said:

Possibly. I'd suggest that "fantastic" is a slight deviation into the world of hyperbole. We certainly weren't abject though.

But this is the whole point of xG: to test whether what you perceived with your eyes, and what the table told you - that we were, at times, fantastic - is borne out in the hard light of statistical analysis. 

Generally it wasn't.

Afobe got injured after the Boro game right? Up to that point we were on a cumulative xG of -0.82, averaging -0.14 per game. After that game we averaged -0.35 until Afobe returned against Blackburn, thereafter we averaged -0.1. So yes by the looks of it we were better when Afobe was available (he didn't play every game post-lockdown, and we had Wells in those matches)...but we were a long way from fantastic. 

Honestly when it comes to Afobe we are dealing with such a small sample as to make the stats almost meaningless...but would Afobe have maintained 1 goal every 2 games across the season and fired us into the top 6...I wouldn't have (and didn't) bet on it.

[Mr Popodopolous]

The 3-5-2/3-4-1-2 surely helped us structurally, and yes we were good though.

I still personally had reservations about a number of areas though even when winning- the lack of possession although if that's part of the game plan, no matter! The chances conceded but this was in part covered by the condensing of space- but lack of possession can contribute to this- and the conversion rate though quite strong, I did wonder how long it could have been kept up- but a fit Afobe for a season and who knows! Guess my reservations at this time and after the strong second wind from late December to mid February, especially the latter, guess these came from building from a position of strength or wanting us to- doing well results wise but questionably performance wise- when doing well and confidence rising make some tweaks to solidify things! LJ in his final season never really seemed so good at this.

Afobe's strong conversion rate may have seen us push on too.

It is also worth noting in mitigation, now I'm thinking back to counter this that the number of injuries to KEY players that we had early on was ludicrous- we had DaSIlva out after Leeds, with Rowe at LWB for half a season, we had Nagy get injured halfway through or thereabouts his 2nd game- and then injured again on International duty out for 2-3 months! Afobe of course, Kalas from Hull away for 2-3 months- in a sense or by one measure with such recurring, rolling problems we were overachieving irrespective of stats and xG!

Those injuries are just off the top of my head sure there were further important ones I've forgotten? A fully fit  Kalas, DaSilva Nagy or Smith and Afobe- as the team gelled together, may well have seen our 3-4-1-2 not only continue to flourish and grow but also our numbers in all areas stabilise and improve over time! Went into the season (I think) with Smith injured too- at minimum that's 4 starters unavailable for the medium to long term, with one first reserve not far behind.

[IAmNick]

2 hours ago, JonDolman said:

Yeah we 'dominated without possession' as LJ said.

I think I remember someone telling me Derby had a better XG than us that night.

But when watching the game we could all clearly see we looked the better side.

Maybe should have won by a more comfortable score line with how comfortable we looked and how many counter attacking opportunities we had as they were pushing for goals.

The problem is when XG gives a false impression of one game. It can do so for many games.

It very often will give a false impression of one game, but that's totally expected and not how it should be used. It's used to analyse trends and patterns over a large number of games, and then it becomes much more useful and interesting.

It's like a doctor saying "Well the stats say on average I see two people with diabetes per day. I haven't seen any today, so these last two folk who have broken fingers must have diabetes or the model is wrong!"

As with all statistics of this type it should be used in conjunction with other data, and not on it's own to form a conclusion - and it being "wrong" isn't necessarily a bad thing. That can itself be an interesting discussion point, or  provide insight into teams who are performing in an unusual way which the model might not account for... and that's what these are for really for the majority of us. To provide insight and provoke discussion.

I often walk away from a game (or season!) with an impression of how we played in my head, but I'm also aware I have a lot of biases depending on our formation, who got picked, who we're playing against - whether we were the better team for the last period or the game/season of the first and so on. These kind of stats can also be interesting to reveal bias in ourselves.

[Davefevs]

2 hours ago, JonDolman said:

Yeah we 'dominated without possession' as LJ said.

I think I remember someone telling me Derby had a better XG than us that night.

But when watching the game we could all clearly see we looked the better side.

Maybe should have won by a more comfortable score line with how comfortable we looked and how many counter attacking opportunities we had as they were pushing for goals.

The problem is when XG gives a false impression of one game. It can do so for many games.

Yep, well explained below by Nick.

37 minutes ago, IAmNick said:

It very often will give a false impression of one game, but that's totally expected and not how it should be used. It's used to analyse trends and patterns over a large number of games, and then it becomes much more useful and interesting.

It's like a doctor saying "Well the stats say on average I see two people with diabetes per day. I haven't seen any today, so these last two folk who have broken fingers must have diabetes or the model is wrong!"

As with all statistics of this type it should be used in conjunction with other data, and not on it's own to form a conclusion - and it being "wrong" isn't necessarily a bad thing. That can itself be an interesting discussion point, or  provide insight into teams who are performing in an unusual way which the model might not account for... and that's what these are for really for the majority of us. To provide insight and provoke discussion.

I often walk away from a game (or season!) with an impression of how we played in my head, but I'm also aware I have a lot of biases depending on our formation, who got picked, who we're playing against - whether we were the better team for the last period or the game/season of the first and so on. These kind of stats can also be interesting to reveal bias in ourselves.

Yep, I tend to look at xG charts in a game for the flow of chances across the 90 minutes, or for challenging the comment “that’s a gilt edged chance” when xG says it’s a 1 in 4.