One stat to rule them all (xG/xA)

[Gambit]

Interesting article about the accuracy of xG. As the author mentions at the beginning we are probably going through chaos stage at the moment, with multiple models available and lack of standardized approach.

http://www.aquestionoftalent.com/data_s ... stigation/

As expected there is a growing market within the game (coaches, managers) that are looking for insights from xG. This makes me think that there will be (or is happening already) a lot of investment available and soon enough, with current technological advances the most superior model will emerge. Already it seems that over the long period of time and even a moderate number of events (i.e.: shots per player) the data is pretty accurate.

Lastly, I recently read about some analytical insights in NBA that changed the way some team play with a significant increase of 3 point shots. I wonder if this means that in few years we will see the decrease of long-range shots in football as those shots success ratio is extremely low.

would there need to be technological advances for that though

I can assure you that nothing annoys a manager, team-mates and fans, more than seeing a player (usually a defender) take a wild 30 yard shot with their weaker foot when a ball breaks from the box!

It's nothing to do with data analysis and everything to do with controlling the rush of blood to the head

ave it!

[konstro]

Interesting article about the accuracy of xG. As the author mentions at the beginning we are probably going through chaos stage at the moment, with multiple models available and lack of standardized approach.

http://www.aquestionoftalent.com/data_s ... stigation/

As expected there is a growing market within the game (coaches, managers) that are looking for insights from xG. This makes me think that there will be (or is happening already) a lot of investment available and soon enough, with current technological advances the most superior model will emerge. Already it seems that over the long period of time and even a moderate number of events (i.e.: shots per player) the data is pretty accurate.

Lastly, I recently read about some analytical insights in NBA that changed the way some team play with a significant increase of 3 point shots. I wonder if this means that in few years we will see the decrease of long-range shots in football as those shots success ratio is extremely low.

would there need to be technological advances for that though

I can assure you that nothing annoys a manager, team-mates and fans, more than seeing a player (usually a defender) take a wild 30 yard shot with their weaker foot when a ball breaks from the box!

It's nothing to do with data analysis and everything to do with controlling the rush of blood to the head

ave it!

I still remember Lovren shot in added time in FA Cup semifinal vs. Aston Villa. Must have been 14/15 season.

I would disagree that it has nothing to do with analytics.Your point is that everyone knows it. Yet, without data it is only an opinion We get insights through data and based on that coach the behaviour that gives us a higher percentage of chances to succeed. Since 3-pointers were introduced in basketball 50 years ago(?) everybody knew that you will get 50% more if you shoot from distance, yet only recently some teams went from 3-pointers being about 25% of shots to 40%. (I have no clue about NBA, this is based on this article I read). Long distance shots are glorified in football as a special skill, yet they make no sense from the statistical point of view as about 2% of them are successful. Here we are talking about the highest level of playing field and supreme athletes that PL players are.

2 situations from last week. MU - Burnley game, first half, United trailing 1-0, Burnley absorbing pressure. Luke Shaw gets a run at Burnley goal, on his favourite left foot and lets it fly from 25 or so metres. Result: Easy save for Pope. Another situation, Arsenal - Liverpool, Second half, Granit Xhaka gets into good position about 30 metres from goal and shoots. He gets lucky and scores, mostly due to Mignolet ineptitude. Although the outcomes are extremely different both events have the same xG of 0.02. And Xhaka is this audacious risk-taker who has the gut to try against the odds. I`m wondering if in the future given the same situation the behaviour might change based on data that is being collected. Obviously, regardless of data, I don`t believe that in this particular instance Xhaka will alter his behaviour. He takes wild swipes in each game and already succeeded 2 times this season so why would he? Is he actively discouraged to take shots from distance?

I`m thinking in more general terms. Like, Liverpool had more than 30 shots vs. Burnley, in September yet the xG was only 2.19. Would it not pay to be more patient instead of shooting at a sight of goal? 30 shots in a single game look impressive but actually is not; given that vast majority of them had no chance to succeed.

Anyway, shots from distance are only one of the possible areas where the use of data could prove to be useful in altering behaviour. I for one really am looking forward to see if football will find it`s counterpart to baseball`s "walking the base, instead of stealing the base" (again, no clue about baseball. Observation based on Moneyball) behavioural change. Or maybe as the second quote in the article indicates data will never get the hold on football decision making and as I write is there are and forever will be thousands of little Granits around the world practising long-distance shots despite little chance to succeed.

[Desperado]

Great stuff Konstro. I think that many of us fisoers are stats geeks and therefore these xG/xA models pull the right strings.

I definitely think that different models are increasingly affecting how the teams play nowadays.



This is from Colin Trainors Twitter in the end of last season. The slide is pretty evident. I also recall that during Guardiola's time at Barcelona there was lengthy discussion about his vision to reduce the low probability finishing attempts. Couldn't find it now but I will look further with time. This hypothesis does indeed pass the eye test with the Man City 17/18.

Your references to NBA and MLB are good and stat based recruitment methods (Moneyball idea) have already been used among European football for example FC Midtjylland and Brentford.

The idea that high proportion of scouting is still conducted by a person watching few matches and after his subjective report are teams ready to splash money around. I'm sure there's still a lot to learn about statistical modeling with in the football community even though it is getting more influence every year. There will be many vacancies to be filled in this sector around Europe and I'm pretty sure that here at Fiso there is many users who'd be able to make a work out of that with a bit of luck and effort.

[pascalevans]

Interesting discussion. All a bit over my head but I seem to remember Sam Allardyce at Bolton instructing his players not to shoot from certain areas. Even he though would allow Okocha to take speculative shots.
If inferior opposition know that you are never going to shoot from over 30 yards then they can defend deeper and make it harder for the other team to have good possession near the area. When playing against players like Coquelin with a weak strike the team can reorganise itself. Xhaka at Arsenal is useful as he creates that uncertainty. By being closed down more it creates more space for the other players. A few missed shots is worth the uncertainty that is generated. It would be interesting to see if Man City have slightly increased their long shooting this season to create uncertainty. It feels as if they have. In chess you have to play the best move and one can be dispassionate but football is more affected by passion and I suspect it would be hard to get footballers never to close down players at a certain distance. They would also get a lot of stick from the crowd, as was the case during Newcastle Man City. I thought Benitez increased the probability of getting a point by effectively making it a 15 minute match at the end. If he did that against all superior teams no one would watch them though.

[blahblah]

Stuff like this reminds me of POMO and the death of English football....

[Gambit]

I would disagree that it has nothing to do with analytics.Your point is that everyone knows it. Yet, without data it is only an opinion

no, my point is that every manger in the country is already acutely aware of the wastefulness of speculative long range swings, and will already telling their players in no uncertain that they are not to just blast the ball at goal from 30 yards.

It's insulting to football managers to suggest they need somebody to give them a spread sheet saying "look, it's probably best your players don't try too many 30 yard shots because only 3% will ever end up in the net"

As mentioned by other people in this thread, each team will likely have players who are encouraged to shoot from outside the box, or at least given license too.

I think you are completely underestimating how much a PL manager/coaches will know about their players and just how in-depth the training sessions and tactical theory can be (especially at the highest level)

As I said, the problem is emotions, you often hear managers bemoaning that players haven't followed instructions or made the wrong decision. Most players know they shouldn't be blasting at goal from 30 yards and just wasting possession, and you often see the manager b0llocking them when they do.

No amount of statistical analysis is likely to change it, the players are not robots.

[konstro]

I would disagree that it has nothing to do with analytics.Your point is that everyone knows it. Yet, without data it is only an opinion

no, my point is that every manger in the country is already acutely aware of the wastefulness of speculative long range swings, and will already telling their players in no uncertain that they are not to just blast the ball at goal from 30 yards.

It's insulting to football managers to suggest they need somebody to give them a spread sheet saying "look, it's probably best your players don't try too many 30 yard shots because only 3% will ever end up in the net"

As mentioned by other people in this thread, each team will likely have players who are encouraged to shoot from outside the box, or at least given license too.

I think you are completely underestimating how much a PL manager/coaches will know about their players and just how in-depth the training sessions and tactical theory can be (especially at the highest level)

As I said, the problem is emotions, you often hear managers bemoaning that players haven't followed instructions or made the wrong decision. Most players know they shouldn't be blasting at goal from 30 yards and just wasting possession, and you often see the manager b0llocking them when they do.

No amount of statistical analysis is likely to change it, the players are not robots.

As the graph presented by Desperado shows it already happened and in all 5 leagues the distance that shots are being taken from have decreased by 5 to 10% in last few years. Obviously, it does not mean they will disappear but with the time I would expect further decrease because they are less economical. Most of the clubs have sport science departments and the use of analytics will grow. The era of big data and increased data accessibility is changing the way we think about analytics. My main question is: what will be the huge change in the behaviour (similar to what happened in baseball and NBA) that will come due to increased use of analytics?
For the record, I`m not trying to insult anyone. I`m the layman here who tries to improve his chances for a good result in a game. I just believe that there are no stupid questions and I`m happy to ask any that will help me to understand overall picture better.

[Stemania]

Great few posts above. I especially like the article linked by konstro, excellent read.

[Data_blade]

Unreal write up there! I myself am very new to data science current of studying through open uni and have managed to produce a rather basic xG model with aid of open source data from various websites. xA interests me more though, I cannot fathom how to make a model or formula to churn out those numbers, does anyone have a link to an article or site that cites an example of formula to workout xA? Trying to currently complete a write up on EFL L1 and L2 with data I have! Also worth noting that companies like stratabet now have secondary xA data the pass before the assist so you can see who is playing those passes from deeper positions if you don't already follow lucid analytics on Twitter. Unreal subjext matter get posted.

[Mav3rick]

Nice twitter account that (@lucid_analytics) and one I've just started to follow, thanks Data_blade.

[gallus]

Unreal write up there! I myself am very new to data science current of studying through open uni and have managed to produce a rather basic xG model with aid of open source data from various websites. xA interests me more though, I cannot fathom how to make a model or formula to churn out those numbers, does anyone have a link to an article or site that cites an example of formula to workout xA? Trying to currently complete a write up on EFL L1 and L2 with data I have! Also worth noting that companies like stratabet now have secondary xA data the pass before the assist so you can see who is playing those passes from deeper positions if you don't already follow lucid analytics on Twitter. Unreal subjext matter get posted.

My guess would be that they simply take the xG numbers and look who made the final pass. Nothing fancier than that. That's how I would do it anyway.

[Cars and Mané]

Thought this would be the best place to post this.

Anyone got a clue what's going on with Huddersfield's expected goals conceded?

They've conceded 12 more than their xgc, the most in the league. Just random variance? Would suggest Newcastle's defence is far worse.

[Stemania]

I had a quick play around with FFS' numbers vs understat's and (on xG at least) they reassuringly seem to generally agree
- maybe surprising considering that they use as far as we understand very different methods of calculation? Good news though.

Now we've had a year of widespread use, I updated the graphs from earlier out of interest in how the two main sources of xG and xA compare - so stats from understat (free) and FFS members area (less free). Unfortunately I couldn't directly compare to Paul Riley's model again due to his stats being 'open play' only.

The grey lines are how many goals and assists the player actually got over the entirity of last season. Blue is understat, red is FFS.



It seems both sources remain reassuringly similar on xG calculation (actually surprisingly closely aligned, to me anyway!), with the same players 'under-' or 'over-'performing the site's respective xG stats over the 17-18 season.



The xA stat seems much trickier. What's immediately noticeable is that the xA stat seems to on average consistantly underestimate the actual number of assists. Certainly compared to how well the xG does anyway. Perhaps there could be reasons for this - that the players usually taking shots are strikers (the best shooters), but passes may be more likely to come from anywhere (striker, winger, center mid - so not necessarily the best passers), making the real stars easily overperform the 'average' pass quality. Or data collection for xA just isn't as refined yet. Who knows, might be something interesting to explore. Either way, despite a very rough correspondence between who under/over-performed according to both sites, there's clearly less of an agreement where modelling assists is concerned. Perhaps understat's are slightly more correlated to reality, compared to FFS.

So atm, it seems fair to say that xG is the more refined stat, and perhaps the most useful of the two if making projections?

[Patrician]

Very cool graph, any chance you could link the spreadsheet?

[Stemania]

Very cool graph, any chance you could link the spreadsheet?

Can pm it, but since the FFS numbers are paid for data probs best not share the whole detailed list publicly.

I realize it would have been useful to share the Key to the graphs with all the players names in case anyone wishes to look them up.

xG graph - sorted by most actual goals

1 Mohamed Salah 32
2 Harry Kane 30
3 Sergio Agüero 21
4 Jamie Vardy 20
5 Raheem Sterling 18
6 Romelu Lukaku 16
7 Roberto Firmino 15
8 Alexandre Lacazette 14
9 Gabriel Jesus 13
10 Glenn Murray 12
11 Son Heung-Min 12
12 Eden Hazard 12
13 Riyad Mahrez 12
14 Álvaro Morata 11
15 Marko Arnautovic 11
16 Sadio Mané 10
17 Pierre-Emerick Aubameyang 10
18 Wayne Rooney 10
19 Luka Milivojevic 10
20 Chris Wood 10
21 Christian Eriksen 10
22 Leroy Sané 10
23 Alexis Sánchez 9
24 Dele Alli 9
25 Wilfried Zaha 9
26 David Silva 9
27 Ashley Barnes 9
28 Anthony Martial 9
29 Jesse Lingard 8
30 Callum Wilson 8
31 Oumar Niasse 8
32 Ayoze Pérez 8
33 Joshua King 8
34 Kevin De Bruyne 8
35 Chicharito 8
36 Xherdan Shaqiri 8
37 Charlie Austin 7
38 Jay Rodriguez 7
39 Olivier Giroud 7
40 Salomón Rondón 7
41 Jordan Ayew 7
42 Aaron Ramsey 7
43 Steve Mounie 7
44 Marcus Rashford 7
45 Marcos Alonso 7
46 Philippe Coutinho 7
47 Pascal Groß 7
48 Abdoulaye Doucouré 7
49 Dwight Gayle 6
50 Paul Pogba 6
51 Shinji Okazaki 6
52 Mame Biram Diouf 6
53 Willian 6
54 Dusan Tadic 6
55 Bernardo Silva 6
56 Laurent Depoitre 6


xA graph - sorted by most actual assists

1 Kevin De Bruyne 16
2 Leroy Sané 15
3 Raheem Sterling 11
4 David Silva 11
5 Mohamed Salah 10
6 Riyad Mahrez 10
7 Christian Eriksen 10
8 Dele Alli 10
9 Paul Pogba 10
10 Henrikh Mkhitaryan 9
11 Aaron Ramsey 8
12 Pascal Groß 8
13 Mesut Özil 8
14 Johann Berg Gudmundsson 8
15 Romelu Lukaku 7
16 Roberto Firmino 7
17 Sadio Mané 7
18 Xherdan Shaqiri 7
19 Willian 7
20 Alex Oxlade-Chamberlain 7
21 Andros Townsend 7
22 Marc Albrighton 7
23 Granit Xhaka 7
24 Aaron Cresswell 7
25 Chris Brunt 7
26 Sergio Agüero 6
27 Son Heung-Min 6
28 Álvaro Morata 6
29 Marko Arnautovic 6
30 Alexis Sánchez 6
31 Philippe Coutinho 6
32 Manuel Lanzini 6
33 Dominic Calvert-Lewin 6
34 Jordon Ibe 6
35 Joe Allen 6
36 César Azpilicueta 6
37 Ben Davies 6
38 Kyle Walker 6
39 Anthony Martial 5
40 Jesse Lingard 5
41 Ayoze Pérez 5
42 Marcus Rashford 5
43 Eric Maxim Choupo-Moting 5
44 Christian Benteke 5
45 Alex Iwobi 5
46 Juan Mata 5
47 Matt Ritchie 5
48 Cesc Fabregas 5
49 Andrew Surman 5
50 Andrew Robertson 5
51 Kieran Trippier 5

[blahblah]

Well there goes a fun quiz

[Patrician]

Here is a graph of the distribution of what % players over and under perform vs xG per 90 minutes. Not quite the distribution I had assumed. I thought it would be more of a normal distribution centred around 0.

SnipImage.JPG

It does seem that 30% overperformance is normal for a top player.

Here is the list of players in the analysis.

[Patrician]

I like to hunt for bargains amongst the players who have under performed, where I assume they will regress to the norm. So players to watch:

Mane
Morata
Sanchez
Pogba
Alli

Players who massively over-performed include:

Gros
Mahrez
Sane
Tosun

[No Way Jose]

Good stuff again mate. Alli is an interesting one. I picked him several times last season expecting him to regress to his mean but he never did and picking him over eriksen burnt me. Won’t be picking him anytime soon!

Mane, Pogba and Sanchez though are another story

[blahblah]

Here is a graph of the distribution of what % players over and under perform vs xG per 90 minutes. Not quite the distribution I had assumed. I thought it would be more of a normal distribution centred around 0.

Nah, that was the same idea in Vid's Comp about the average being mid table?

[Stemania]

Interesting stuff! Perhaps it says a lot about the quality of the PL!

[Droughton]

Yes lots of data which is highly useful.

I would just point out that if you do not know the root cause reasoning of why players are over/under performing to partner the data then it is just data.

If a player underperformed to expected goals last season doesn't mean they will bounce back this year by 'averages'. It may just be that they have lost that clinical edge or that the chances when you watch them aren't actually that clear cut.

[Ironfist]

Yes lots of data which is highly useful.

I would just point out that if you do not know the root cause reasoning of why players are over/under performing to partner the data then it is just data.

If a player underperformed to expected goals last season doesn't mean they will bounce back this year by 'averages'. It may just be that they have lost that clinical edge or that the chances when you watch them aren't actually that clear cut.

... or that the data is only interperted in a way to justify this new 'stat to rule them all' and people forget that in sports Tymora reigns Supreme.

[Droughton]

... or that the data is only interperted in a way to justify this new 'stat to rule them all' and people forget that in sports Tymora reigns Supreme.

True!

[Stemania]

Here is a graph of the distribution of what % players over and under perform vs xG per 90 minutes.

Nice article here analysing how individuals have done against the xG benchmark in recent years - based on understat numbers:

http://eightyfivepoints.blogspot.com/20 ... goals.html

The idea being to view it as a "how do we decide who is actually good at finishing" metric.

[baganboy]

Very interesting. Thanks for sharing.

[Ardrageen]

I'm a bit surprised by Aguero's performance. I thought he would rate a little higher. No surprises at the bottom. Poor Benteke

[Droughton]

Thanks Stem

[blahblah]

Given that xG is calibrated based on the success rate of shots from mostly inferior players, we should therefore expect elite forwards, such as Kane, to outscore the xG total of their shots over a season. But to what extent?

Exactly my point about this, yippeeeee... It is just a measure of the quality of chances created for that player.

I'm a bit surprised by Aguero's performance. I thought he would rate a little higher. No surprises at the bottom. Poor Benteke

Townsend is a fair bit higher than I would expect, as he cuts in for the right then boots it with his left foot hoping to curl it near the goal, which got me thinking.... Maybe some of the high ones are because they finish more of those Townsend 1 in 10 chances? This fits in with Defoe down at 1, as he (all the following will be generalisaions, so no random exemptions please ) essentially gets tap ins and one on ones as opposed to Eden, who doesn't. In other words the average player can\does score the tap ins (but doesn't get there) but not the others, and I would use Kane as an example for one of his early seasons and being well over xG as he got a lot from outside the box.

re Kun, maybe mixing up his shots with both feet helps keep the GK guessing when he is going to shoot?

Also headers would be better out of this, imho, as a separate grouping.

[fred1266]

is this method still reliable?