Match Stats - Data Perspective

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
I decided to get a closer look at Match Stats just for my own interests, but migh as well post it here if anyone is interested. I will try to do it for the next couple of weeks and see if we can get some interesting conclusions which are contradiction to common knowledge - which data analysis is for.

THE POINT OF THIS THREAD is to confront the popular views which are considered as "obvious" with data and compare those phenomenons with other teams/ different combinations/ different players. Please restrain from posting cliché like "I don't need stats to see we're shit/ the only thing that matters is the league table" - this is not why we're here.

I am basing my analysis on WhoScored (players stats, heatmaps etc) and Understat (xG/xGA Timing Chart) mostly. Will keep OP updated if I find something new and interesing. Most of the general stats I take from fbref.

What is xG?
Expected goals (xG) measures the quality of a shot based on several variables such as assist type, shot angle and distance from goal, whether it was a headed shot and whether it was defined as a big chance. Expected assists (xA) measures the likelihood that a given pass will become a goal assist. It considers several factors including the type of pass, pass end-point and length of pass.
If you see that chance is described as having an xG rating of 0.1 that means a player would be expected to score from the chance 10per cent of the time - a one in ten chances (for example a header from a corner). If a chance is described as 0.5xG it should be scored 50% of the time (for example a easy one on one chance).

If you sum those parameters for each goalscoring situation, you get a Timing Chart. The cummulative value will be the final xG - the higher the more likely team is to score goals. Depending on final xG for opposition team, the delta should be a good indicator of what would be the score, not including converting chances parameter.

This graphs shows that our best period came in the second half, when WBA couldn't create anything.

So for example, lets compare two extreme cases - West Bromich Albion game and Brighton in EPL.
We've won WBA game and I thought we were quite unlucky not to score more goals. But it doesn't seem like a popular reaction on Redcafe as most people are moaning about lucky win. So was it a lucky win? We've had a Total xG score 2,43. So lets assume if we played that game 100 times, the average goals scored would be 2,43. At the same time, xG for WBA was at the level of 0,44, which means we would win most games by 2 goals. Based on that data, it seems we were quite unlucky not win by a higher margin. In the end we've scored only one goal (penalty), which can be explained by worst conversion rate in the EPL.

Now lets see the Brighton game, which we've won 3-2. I thought we were extremely lucky to get away with any point from that game, so I looked up the stats and guess what - we got a total xG score of 1,58 compared to Brighton 2,98. So basing on the number and quality of chances we've got, we should be losing that game by at least one goal (graph below). We've also made a total number of 3 shots on target (scored 3), compared to 7 vs WBA. If we played that game multiple times, we would concede 3 goals on average so quite lucky on that end too.
 
Last edited:
  • Like
Reactions: Invictus

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Manchester United 1 - 0 West Bromwich Albion, Fernandes Penalty
21/11/20, xG=2,43, xGA=0,44
  • Shots 17 (7 on target) vs 7 (2 on target)
  • Possession 64% vs 36%
  • 663 passes with 84% accuracy vs 385 passes with 77% accuracy

Comparing to WBA vs Tottenham game:
  • Shots 19 (T) vs 12 (WBA), on target 5 vs 2
  • Possession 59% (T) vs 41% (WBA)
  • 508 passes with 78% pass accuracy for Tottenham vs 340 passes with 66% accuracy for WBA
Kane scored great header in 88', 0-1 final score. We created better chances (compare Spurs vs United). We had better chances than Tottenham based on xG measure (2,43 vs 1,87), and the chances West Bromich Albion created vs us were less dangerous (0,44 vs 0,8 against Spurs). We also got higher score on expected points parameter. WBA keeper was their highest rated player, while vs Tottenham he was 7th most rated player. Maybe that will change perspective of some people.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
I've been analyzing yesterday game vs WBA, one thing that is quite obvious is how heavy is our left wing compared to right wing. It's common knowledge but just take a look at the heatmaps:
Bruno, Rashford, Martial separately


Bruno, Rashford, Martial combined - they are all occupying similar space. Even if they take positions on the right, it's very deep and with little attacking intent.


Wide attackers and fullbacks combined, you see our right wing is not even getting close to the box


Last pic - All front 4 combined. Wan Bissaka and midfield adds nothing to the right side while attacking. On the left Telles should be included but that would blur an image.


Now I see a lot of people criticizing midfield for not being offensive enough. However, I believe what we should focus on how to bring balance back between left and right wing. It's very easy for opposition to congest center and left side while defending, because there is no threat coming from the right, and if we want to improve, we need to address this issue first and at least try to stretch opponent defense formation.

Possible solutions:
  • start Cavani at CF, and shift one of Martial/Rashford to the RW. It's not ideal as neither of them seems comfortable playing there, but it might be beneficial for the team. Both of them are struggling to put the ball into the net, so not a big loss.
  • wait for Greenwood (whatever is happening), and play him from the right. He likes to roam around, and is left footed but he seems like a threat wherever he plays. His defensive contribution is weak so it's 1to1 Mata replacement.
  • switch to back 3 with McTominay as RCB, play Fred and van de Beek in front of the 3 defenders. Donny seems quite comfortable on the right, unlike Bruno & Pogba he doesn't have natural inclination to move to the left. Makes no sense to play him in midfield two as it'd limit his license to push forward. This would be my preferred approach as we have one too many attacking midfielder.
  • survive until winter transfer window and splash the cash for a proper right winger. This is not happening considering the way we do business.
 

Mick1

Full Member
Joined
Feb 17, 2014
Messages
407
Very interesting.
Can you please provide based on your stats indivudual xg for our players against said opposition?
Also, how does winning penalties or pre penalty pass and the even pre penalty free quick, maybe with weighted xgs per free kick/penalty per player?
For examle for Fernande shooting a penalty, we account for Fernendes' conversion as 0.92 xg/penalty, updated on an incremental basis, xg per player for Fernandes 0.12 and for the winner of the penalty as 0.8 ( the average in the league) since he was theoretically in a position to get a goal. Hence Rashord for example winning a penalty would account in his xg as 0.8, whilst Fernandes as 0.12. Breaking this down would show a more complete view, with a bias.
I havent been able to find a source online, and too lazy to do it myself.
 

Gabagoo

New Member
Newbie
Joined
Oct 15, 2020
Messages
246
So the steps involved in this analysis are (please correct me if I'm wrong):
  1. Source and collect the data (obviously knowing the xG model is first necessary to correctly read the data)
  2. Data --> cumulative sums over time per team --> time graphs
  3. Delta at max time suggests the mean average final score in a theoretical series of matches
We can surmise from looking at all matches for all teams in a season, what results, total points and positions each team 'should' take.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Very interesting.
Can you please provide based on your stats indivudual xg for our players against said opposition?
Also, how does winning penalties or pre penalty pass and the even pre penalty free quick, maybe with weighted xgs per free kick/penalty per player?
For examle for Fernande shooting a penalty, we account for Fernendes' conversion as 0.92 xg/penalty, updated on an incremental basis, xg per player for Fernandes 0.12 and for the winner of the penalty as 0.8 ( the average in the league) since he was theoretically in a position to get a goal. Hence Rashord for example winning a penalty would account in his xg as 0.8, whilst Fernandes as 0.12. Breaking this down would show a more complete view, with a bias.
I havent been able to find a source online, and too lazy to do it myself.
Thanks, will take a look on that. I've only started this today so will try to look it up.

I've been there, just not sure how to use that as I can't find data for any particular game (I'll be building trend graphs).
 

P-Nut

fan of well-known French footballer Fabinho
Joined
Oct 16, 2015
Messages
21,602
Location
Oldham, Greater Manchester
Looks interesting although the 2 game s you've picked I'd say most people would say you could tell what the stats would be without looking.

I'm always interested in those games that defy the stats, where it feels like you've played well, but the stats say it was a tight game or you should have lost.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Looks interesting although the 2 game s you've picked I'd say most people would say you could tell what the stats would be without looking.

I'm always interested in those games that defy the stats, where it feels like you've played well, but the stats say it was a tight game or you should have lost.
Yeah I just wanted to check if the stats would confirm my prediction, and thought it was a good example to explain how it works.

Anyway, here's the full data for this season for EPL:

Clearly the blue line is a good indication of final outcome, means it's almost always on the side of winning team - again, Brighton being the exception. We were really lucky that game.
I think this is actually a good measure to monitor, the calculation is on the safe side but pointing the right way.

I was actually quite surprised how accurate the expected score is to the real score. It shows we've overachieved in Brighton game and Newcastle (although we were expected to win that one anyway), and we underachieved in Tottenham (again, would've lost anyway) and recent WBA game. Seems reasonable, I will keep it updated.

One thing is we're clearly improving since Newcastle game, which is when we dropped Pogba and Matic. Since then we're on 1,5 xG diff average. I would like to update this with Champions League games but can't find those game stats. Maybe I could add conversion rate if I can find it easy accessible.
 
Last edited:

SinNombre

Full Member
Joined
Jan 26, 2019
Messages
2,626
Well the key takeaway should be that we are playing at par since the Newcastle game where Ole started playing Fred and Mct more consistently as the double pivot.

I generally assume that xG +/- 0.75 should generally end in a draw so we should have been on a run of W,D,D,W,W which would have been 11 points in the past 5 games. I think that is about par for this team.

We obviously still need that right sided attacker, and having Mason back will help.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Well the key takeaway should be that we are playing at par since the Newcastle game where Ole started playing Fred and Mct more consistently as the double pivot.

I generally assume that xG +/- 0.75 should generally end in a draw so we should have been on a run of W,D,D,W,W which would have been 11 points in the past 5 games. I think that is about par for this team.

We obviously still need that right sided attacker, and having Mason back will help.
That's my understanding as well, but we'll know more as the season progresses. But we should rather focus on improving finishing, because we're on the right side of chances created - chances conceded.
 

Lentwood

Full Member
Joined
Jan 2, 2015
Messages
6,783
Location
West Didsbury, Manchester
Fans really struggle to understand probability in relation to football matches because they are emotionally invested and often affected by the equally emotional and often false narratives offered by pundits.

Quite often, pundits push a narrative that a top team have been “poor” or “underwhelming” if they don’t run out 3 or 4 goal winners against weaker sides but there’s a huge, huge gap between ‘very good’ and ‘poor’ which is never really explored. You never see a commentator say “Utd have been very slightly above average today” or “Utd have played at roughly their level and so have West Brom”.
 

UNITED ACADEMY

New Member
Joined
Aug 14, 2018
Messages
13,127
Supports
Erik ten Hag
Manchester United 1 - 0 West Bromwich Albion, Fernandes Penalty
21/11/20, xG=2,43, xGA=0,44
  • Shots 17 (7 on target) vs 7 (2 on target)
  • Possession 64% vs 36%
  • 663 passes with 84% accuracy vs 385 passes with 77% accuracy

Comparing to WBA vs Tottenham game:
  • Shots 19 (T) vs 12 (WBA), on target 5 vs 2
  • Possession 59% (T) vs 41% (WBA)
  • 508 passes with 78% pass accuracy for Tottenham vs 340 passes with 66% accuracy for WBA
Kane scored great header in 88', 0-1 final score. We created better chances (compare Spurs vs United). We had better chances than Tottenham based on xG measure (2,43 vs 1,87), and the chances West Bromich Albion created vs us were less dangerous (0,44 vs 0,8 against Spurs). We also got higher score on expected points parameter. WBA keeper was their highest rated player, while vs Tottenham he was 7th most rated player. Maybe that will change perspective of some people.
Nice work with the stats. I also thought the same thing before you posted it, the chances we created were actually chances that were really good. Sam Johnstone actually had worldie. I'm sure he was the motm.

Better front three or clinical finisher would have finished that game with big score. And I remember the Spurs game vs West Brom, it was much worse than our game. Their goal being their highest xG and still lower xG than majority our chances and ironically it was actually from Sam's blunder. If we swap Martial & Mata with Kane & Son yesterday, people will have different perspective, those two are more clinical and more creative than our front three.
 

Classical Mechanic

Full Member
Joined
Aug 25, 2014
Messages
35,216
Location
xG Zombie Nation
Well the key takeaway should be that we are playing at par since the Newcastle game where Ole started playing Fred and Mct more consistently as the double pivot.

I generally assume that xG +/- 0.75 should generally end in a draw so we should have been on a run of W,D,D,W,W which would have been 11 points in the past 5 games. I think that is about par for this team.

We obviously still need that right sided attacker, and having Mason back will help.
That tolerance is too high for me. Penalties are 0.75xG.
 

Oranges038

Full Member
Joined
Oct 19, 2020
Messages
11,783
Personally I think Xg is a nonsense way to analyse individual chances. No two chances are ever the exact same.

There are too many variables in football for this to be the basis for an accurate measurement of whether or not each chance should be converted.
 

Leftback99

Might have a bedwetting fetish.
Joined
Jan 11, 2015
Messages
13,966
Good thread. We've been poor Xg wise so far this season (but improving) but many don't appreciate how evenly matched a lot of PL games are.
 

Bilbo

TeaBaggins
Joined
Sep 27, 2004
Messages
14,207
The left side bias in our team has been there for some seasons now, and heat maps always emphasis this. Its really something that we need to address, and for me it hugely influences peoples perceptions of our full backs.

AWB gets regular criticism for being poor offensively, and he isn't great, but he tends to have far fewer options than Shaw would when in possession. He has more space to work with but is often struggling to find a pass.

What it does mean is that he has better options for a cross since our players tend to be attacking the goal from stronger starting positions, but because he has so little support its much harder for him to get into those dangerous crossing positions. On the flipside Shaw frequently finds himself in strong offensive positions but our players are too close to him. Crossing is rarely an option. Its slightly unfortunate that we now have a LB with the attributes that Telles has. He'd be far more useful to us as a RB i feel.

As much as I feel its better for us overall that we signed squad depth rather than go all in on Sancho, the RW position is still a gaping hole in the squad and is a key reason why we sometimes struggle to break down defences. Teams are relatively happy for us to have space on that side and we end up looking congested and trying to pass through the eye of a needle.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Personally I think Xg is a nonsense way to analyse individual chances. No two chances are ever the exact same.

There are too many variables in football for this to be the basis for an accurate measurement of whether or not each chance should be converted.
I don't analyze individual chances, I sum them up and compare created to conceded (xG to xGA) game to game. I want to build trends, but I'll also be able to compare players with larger sample available.
 

Forevergiggs1

Full Member
Joined
Sep 22, 2019
Messages
3,443
Location
Barcelona
Supports
United
Good post and I'm sure you put in a lot of effort but when we need match stats to explain why we just about beat West Brom then that for me says there's a lot bigger problems than some are willing to admit.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Good post and I'm sure you put in a lot of effort but when we need match stats to explain why we just about beat West Brom then that for me says there's a lot bigger problems than some are willing to admit.
Care to explain what do you mean? I'm only trying to tell people not to overreact, because we don't really have any influence on currently biggest issue (converting chances into goals). I'm also showing how we've improved since season has begun - both offensively and defensively.
 

Forevergiggs1

Full Member
Joined
Sep 22, 2019
Messages
3,443
Location
Barcelona
Supports
United
Care to explain what do you mean? I'm only trying to tell people not to overreact, because we don't really have any influence on currently biggest issue (converting chances into goals). I'm also showing how we've improved since season has begun - both offensively and defensively.
I don't think I'm the only one who thinks stats have their uses but the best way to analise a team is by the trusted old method, the eye test and personally I'm seeing little improvement. Great result against PSG. Very negative performance against Chelsea. Great result against Leipzig followed by a terrible performance against Arsenal then an even worse one against the Turks, followed by a good result against Everton and even though yesterday we won it was another terrible performance whatever the stats say. I'm not saying it won't get better but at the minute what you're basically doing is sugar coating a piece of turd.

I can understand your point about if we converted more of our chances we'd obviously have a better league position but I'm sure there's a lot of other clubs can say the same thing. The same as the defensive record improving. I don't need stats to tell me playing with a double pivot against one of the worst teams in the league we're not going to be conceding a lot of goals. Of course there's positives but until a whole lot of kinks have been worked out and we go on some kind of run I'm not getting too excited.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
I don't think I'm the only one who thinks stats have their uses but the best way to analise a team is by the trusted old method, the eye test and personally I'm seeing little improvement. Great result against PSG. Very negative performance against Chelsea. Great result against Leipzig followed by a terrible performance against Arsenal then an even worse one against the Turks, followed by a good result against Everton and even though yesterday we won it was another terrible performance whatever the stats say. I'm not saying it won't get better but at the minute what you're basically doing is sugar coating a piece of turd.

I can understand your point about if we converted more of our chances we'd obviously have a better league position but I'm sure there's a lot of other clubs can say the same thing. The same as the defensive record improving. I don't need stats to tell me playing with a double pivot against one of the worst teams in the league we're not going to be conceding a lot of goals. Of course there's positives but until a whole lot of kinks have been worked out and we go on some kind of run I'm not getting too excited.
You're seeing little improvement compared to what? Start of the season, last game, or last season? Because I can show you the we've improved massively in last 3 games vs first 3 games in total shots, shots on target, passing accuracy, and reduced number of chances and goals conceded. The only thing that has gotten worse is conversion rate (well "worse" doesn't tell the whole story; inconsistent would be more accurate, because sometimes we score 5 from 7 shots).

Anyway, I don't want to persuade you to use data if you're not interested. But I think this adds another dimension to discussion rather than a bar level of "United were very lucky to win because penalty"- and I've seen that thrown yesterday a lot. If you look at the context, we've been lucky this season (against Brighton), but yesterday we've been actually rather unlucky.
 

Oranges038

Full Member
Joined
Oct 19, 2020
Messages
11,783
I don't analyze individual chances, I sum them up and compare created to conceded (xG to xGA) game to game. I want to build trends, but I'll also be able to compare players with larger sample available.
This is not an argument aimed at you or your work in the post, which you obviously spent a lot of time on.

It's just referenced in the post about how it works for chances. I've read up on it before. All it does is put a number on the amount of goals we should have scored.
I just don't see the point of it as a stat or a valid way of discussing the outcome of a game, you either scored more than the other team and won or you didn't.

The crossbar "chance" warrants a 0.03 xG. That means, a shot like that, from that angle and so, is expected to be scored 3 per cent of the time.
I saw this yesterday in reference to Robinson's effort. When you see people reference as a chance as .03 xg as if it means something it just makes me wonder whether these people actually watch and understand the game or whether they just spew out this stuff to make it sound like they know something.

You can tell by watching the game yesterday, that Rashford and Martial should have scored along with the penalty that's 3. West Brom should have had at least 1. Your work tells us the game should gave finished 2.4 to 0.44. You need to able to watch the games and understand when good scoring opportunities present themselves and when players should score.

I just see it as a pointless analytical data point that is of more use to gamblers who arent physically able to watch every game so they can makes bets on goals scored than any one else.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
This is not an argument aimed at you or your work in the post, which you obviously spent a lot of time on.

It's just referenced in the post about how it works for chances. I've read up on it before. All it does is put a number on the amount of goals we should have scored.
I just don't see the point of it as a stat or a valid way of discussing the outcome of a game, you either scored more than the other team and won or you didn't.
I think of it as an indicator if the team is heading the right way. For example, early in the season we were not creating much and conceding massive amount of chances. We've improved on both xG and xGA, and we're more likely to score more than one goal per game. Building a database like that, you're able to see patterns for example when chances created start dropping you might take a look what changed. Was it a opposition who outplayed us (Arsenal), or maybe we were too careful (Chelsea). If you see games when xGA was higher, you might take a look who played in midfield/defense on that day and so on. This is a just a tool, we'll see if something interesting pops out. It takes me 2 minutes to update so no big issue.

You can tell by watching the game yesterday, that Rashford and Martial should have scored along with the penalty that's 3. West Brom should have had at least 1. Your work tells us the game should gave finished 2.4 to 0.44. You need to able to watch the games and understand when good scoring opportunities present themselves and when players should score.

I just see it as a pointless analytical data point that is of more use to gamblers who arent physically able to watch every game so they can makes bets on goals scored than any one else.
Like I said before, basing on those stats if we played this game over and over again, we should've won by 2 goals (3-1 or 2-0 most likely outcome). From what I've seen so far, those xG numbers are underestimated because the goal difference is bigger in most cases. However, it's almost always on the right side (over 0 - more likely to win, below 0 - more likely to lose). So basically if you track this game by game, you should be able to see patterns.

Brighton game is the only exception so far, which means those indicators are fairly accurate. We're showing positive trend on both xG and xGA and that is my point. I refuse to judge game only by score, this is "top level view" but not enough to draw conclusions. You made a good point about gamblers, I wonder how they analyse those.
 

Bilbo

TeaBaggins
Joined
Sep 27, 2004
Messages
14,207
There's nothing wrong with a little analysis @Borys. Good thread. Don't let it die. Win or lose we will always benefit from more analytical discussion.
 

Web of Bissaka

Full Member
Joined
Feb 20, 2017
Messages
8,553
Location
Losing to Comeback Winning!
:)Good thread again, stats is always interesting, but it (i.e. data) shouldn't be the main argument or perspective IMO.
Personally I prefer to have many perspectives then try to combine all, balance it well and so on to get the full picture.

I'm not seeing any contradictions so far of the data presented in OP and so on with what we can already see clearly just from watching the players perform with our eyes.

Do more data analysis of past and upcoming games this season please.
 

Forevergiggs1

Full Member
Joined
Sep 22, 2019
Messages
3,443
Location
Barcelona
Supports
United
You're seeing little improvement compared to what? Start of the season, last game, or last season? Because I can show you the we've improved massively in last 3 games vs first 3 games in total shots, shots on target, passing accuracy, and reduced number of chances and goals conceded. The only thing that has gotten worse is conversion rate (well "worse" doesn't tell the whole story; inconsistent would be more accurate, because sometimes we score 5 from 7 shots).

Anyway, I don't want to persuade you to use data if you're not interested. But I think this adds another dimension to discussion rather than a bar level of "United were very lucky to win because penalty"- and I've seen that thrown yesterday a lot. If you look at the context, we've been lucky this season (against Brighton), but yesterday we've been actually rather unlucky.
Little improvement on Oles tenure as a whole. I do appreciate the hard, laborious work you've put into the OP but I'm more interested in the team as a unit and for me stats don't show the whole picture. Individually we have some very good players who can create more openings, score more goals or assist than the average team and even though what you say is true that we've improved in the last 3 games for me it's too short a period for it to be conclusive proof that we're on the right track.

I'll gladly come back after the Xmas period and compare the last 3 games with the next 11 to see whether or not we've kept improving and if we have I'll talk stats with you but until then I'm afraid I'm sticking with the old trusty eye method as my confirmation.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Little improvement on Oles tenure as a whole. I do appreciate the hard, laborious work you've put into the OP but I'm more interested in the team as a unit and for me stats don't show the whole picture. Individually we have some very good players who can create more openings, score more goals or assist than the average team and even though what you say is true that we've improved in the last 3 games for me it's too short a period for it to be conclusive proof that we're on the right track.

I'll gladly come back after the Xmas period and compare the last 3 games with the next 11 to see whether or not we've kept improving and if we have I'll talk stats with you but until then I'm afraid I'm sticking with the old trusty eye method as my confirmation.
Feel invited mate. Like I said, it's just for fun and my own training with data analysis. I'll be happy if there are some people interested, but I also understand there is a lot of people who are focused on watching games only.

Just remember the memory will somehow twist reality with every passing day.
 

brzez

New Member
Newbie
Joined
Oct 26, 2020
Messages
86
I saw this yesterday in reference to Robinson's effort. When you see people reference as a chance as .03 xg as if it means something it just makes me wonder whether these people actually watch and understand the game or whether they just spew out this stuff to make it sound like they know something.
Well, it’s interesting that someone chooses to quote something but forgets the context. My quote was in direct regards to people that were claiming that we robbed them and didn’t deserve the points. Whereby I pointed out that they had shit chances. An inshallah long shot from 40 yards doesn’t change that. And even if you believe it or not, that type of shot, in that angle and all, ends up in the net 3 per cent of the time. The facts are there...

But you’ve pretty much proved yourself to be clueless, as xG and xGa are a very well established statistics metric that is used even by professional clubs.
 

Web of Bissaka

Full Member
Joined
Feb 20, 2017
Messages
8,553
Location
Losing to Comeback Winning!
Don't let this discourage you. I'm interested in you future posts about the same thing - data analysis.

I agree with your other points, we are improving... but is it enough? no. Accepting it meant we're accepting low standard. But it's nice to see and know that we are improving this season, even if it's a bit and still need seriously many steps. It's obvious we're improving though, just from seeing the games. The same problems daily seasons of before still exist <-- that's what frustrate people and argued we are just not improving overall.

Anyway.
I just don't get your two... "conclusions"?

First one.
So for example, lets compare two extreme cases - West Bromich Albion game and Brighton in EPL.
We've won WBA game and I thought we were quite unlucky not to score more goals. But it doesn't seem like a popular reaction on Redcafe as most people are moaning about lucky win. So was it a lucky win? We've had a Total xG score 2,43. So lets assume if we played that game 100 times, the average goals scored would be 2,43. At the same time, xG for WBA was at the level of 0,44, which means we would win most games by 2 goals. Based on that data, it seems we were quite unlucky not win by a higher margin. In the end we've scored only one goal (penalty), which can be explained by worst conversion rate in the EPL.

Now lets see the Brighton game, which we've won 3-2. I thought we were extremely lucky to get away with any point from that game, so I looked up the stats and guess what - we got a total xG score of 1,58 compared to Brighton 2,98. So basing on the number and quality of chances we've got, we should be losing that game by at least one goal (graph below). We've also made a total number of 3 shots on target (scored 3), compared to 7 vs WBA. If we played that game multiple times, we would concede 3 goals on average so quite lucky on that end too.
Isn't people moaning about it being a lucky win because.. (1) we didn't concede the penalty fouled by Bruno (for me it's a penalty), and the (2)retaking of our penalties which thankfully Bruno scored in the 2nd. Our finishing is so bad, we were basically heading into a draw.

I also have problems with your statement stating using xg that if we play the game again, it'll be what the xg stated. I think this is the problem with people using data as the deciding point in argument - the assumption that data will (1) still be the same and (2) will decide clearly on average of what "should" happen. It doesn't work that way and I'm sure you'll know this or maybe not yet.

A) If we play the game again, won't the stats be different? The problem is this idea that playing the "same" games again will yield the same xg stats. No it won't. The stats will be different as players performing differently eg. if let's say West Brom won the penalty, it'll spark different reactions to their players as does our players, then creating different xg stats. Every sequence of events per "same" game happen before will affect what'll happen later, thereby changing the whole course of events and the xg stats. so every repetitive games will be different and shall have different unique xg stats.

B) Of course, even if you're still arguing with that line of argument that the xg will maintain being consistent, then why only xg that remains consistent? What happen on the pitch and stats-wise conversion rate should also maintain consistent no? I mean if we'll be creating a lot of that chances again if we repeat the same games, then we will also be wasting those chances again in the same repeated games.

So using xg to maybe prove whatever you're trying to prove there is just full of flaws. Actually it's not quite clear what you're trying to prove here. Is it lucks?


Second one.
......

Kane scored great header in 88', 0-1 final score. We created better chances (compare Spurs vs United). We had better chances than Tottenham based on xG measure (2,43 vs 1,87), and the chances West Bromich Albion created vs us were less dangerous (0,44 vs 0,8 against Spurs). We also got higher score on expected points parameter. WBA keeper was their highest rated player, while vs Tottenham he was 7th most rated player. Maybe that will change perspective of some people.
Don't see the point here.
Is it to show that we are better than Spurs when facing the mighty West Brom?

It's two different games where West Brom performed differently - their players performed better vs Spurs, but performed worser vs us except for their keeper who had a average game vs Spurs and "worldly" performance vs us. Then again when opposing team did many shit shootings and the ball goes towards you, then it's easier to save and appear great. Credit to Johnstone for stepping-up against us, and their whole team stepped up vs Spurs.

Also our players and Spurs players performed differently and uniquely as they did for that two different games. It's also on two different days. I'm sure you know about form - there's the team form and then there's the individual players form.

So I still don't get what perspective needed changing with this data.
 

Forevergiggs1

Full Member
Joined
Sep 22, 2019
Messages
3,443
Location
Barcelona
Supports
United
Feel invited mate. Like I said, it's just for fun and my own training with data analysis. I'll be happy if there are some people interested, but I also understand there is a lot of people who are focused on watching games only.

Just remember the memory will somehow twist reality with every passing day.
Unfortunately I think you picked a bad day to open the OP even though you had good intentions. In my experience after a game like yesterday people don't want to hear about stats. Hopefully if we go on a good run people will become more interested because right now most are focused on performance only but that will change if a bit of good feel factor is brought In to it. This should be a fun thread so you can imagine why there aren't many takers at the minute.

Anyway. If it helps you with your data analysis training then you have nothing to lose. Good luck with it and I hope to see you here around Xmas
 

Chipper

Adulterer.
Joined
Oct 25, 2017
Messages
5,464
Well the key takeaway should be that we are playing at par since the Newcastle game where Ole started playing Fred and Mct more consistently as the double pivot.

I generally assume that xG +/- 0.75 should generally end in a draw so we should have been on a run of W,D,D,W,W which would have been 11 points in the past 5 games. I think that is about par for this team.

We obviously still need that right sided attacker, and having Mason back will help.
The xG sites have expected points too, isn't that a better way to do it? I think they plug in the xG and run a Monte Carlo simulation of the match a large number of times and work out how many points on average a team would have got.

Understat has:
Newcastle (A) 2.38
Chelsea (H) 1.73
Arsenal (H) 0.62
Everton (A) 2.38
WBA (H) 2.74
Total: 9.85

I don't get the 0.75 each way on your allowance for a draw. could you explain that? I'd have thought anything under 0.49 goals between the teams would be a draw if using a rough system like that, anything above 0.51 would be a win to the team with the higher xG and anything precisely 0.50 leaves me not knowing what to do!

If one team has 0.51 xG more than the other then don't you round up to 1 as they're more than half a goal better? They're not guaranteed to win it at all let alone by 1 of course, they never are with xG anyway but if you had to call it isn't that what you'd have to pick?

So:
+/- 0.49 = draw
+ 0.51-1.49 = 1 goal win
+ 1.51-2.49 = 2 goal win
+2.51-3.49 = 3 goal win etc.

Would be W, D, L, W, W for the last 5 and 10 points doing it my way which by luck is exactly what happened.

Maybe I'm way off and not realising something obvious.
 
Last edited:

Oranges038

Full Member
Joined
Oct 19, 2020
Messages
11,783
I think of it as an indicator if the team is heading the right way. For example, early in the season we were not creating much and conceding massive amount of chances. We've improved on both xG and xGA, and we're more likely to score more than one goal per game. Building a database like that, you're able to see patterns for example when chances created start dropping you might take a look what changed. Was it a opposition who outplayed us (Arsenal), or maybe we were too careful (Chelsea). If you see games when xGA was higher, you might take a look who played in midfield/defense on that day and so on. This is a just a tool, we'll see if something interesting pops out. It takes me 2 minutes to update so no big issue.


Like I said before, basing on those stats if we played this game over and over again, we should've won by 2 goals (3-1 or 2-0 most likely outcome). From what I've seen so far, those xG numbers are underestimated because the goal difference is bigger in most cases. However, it's almost always on the right side (over 0 - more likely to win, below 0 - more likely to lose). So basically if you track this game by game, you should be able to see patterns.

Brighton game is the only exception so far, which means those indicators are fairly accurate. We're showing positive trend on both xG and xGA and that is my point. I refuse to judge game only by score, this is "top level view" but not enough to draw conclusions. You made a good point about gamblers, I wonder how they analyse those.
Well, it’s interesting that someone chooses to quote something but forgets the context. My quote was in direct regards to people that were claiming that we robbed them and didn’t deserve the points. Whereby I pointed out that they had shit chances. An inshallah long shot from 40 yards doesn’t change that. And even if you believe it or not, that type of shot, in that angle and all, ends up in the net 3 per cent of the time. The facts are there...

But you’ve pretty much proved yourself to be clueless, as xG and xGa are a very well established statistics metric that is used even by professional clubs.
Where are the facts?

I responded to your comment on this yesterday and asked where that figure of .03 came from. I'd like to know, where are the facts that have decided that Robinson will score from that exact scenario 3 out of every 100 times? I didn't see a response yet, so maybe now you can enlighten me.

I know what it is and how's it's used as a metric. Arguing about xg and xa after a game is the same argument as, we should have won or we should have lost its just putting a number on it to try and make it more valid.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Where are the facts?

I responded to your comment on this yesterday and asked where that figure of .03 came from. I'd like to know, where are the facts that have decided that Robinson will score from that exact scenario 3 out of every 100 times? I didn't see a response yet, so maybe now you can enlighten me.

I know what it is and how's it's used as a metric. Arguing about xg and xa after a game is the same argument as, we should have won or we should have lost its just putting a number on it to try and make it more valid.
Let me quote OPTA for you
EXPECTED GOALS (XG)
EXPECTED GOALS (XG) MEASURES THE QUALITY OF A SHOT BASED ON SEVERAL VARIABLES SUCH AS ASSIST TYPE, SHOT ANGLE AND DISTANCE FROM GOAL, WHETHER IT WAS A HEADED SHOT AND WHETHER IT WAS DEFINED AS A BIG CHANCE.

ADDING UP A PLAYER OR TEAM’S EXPECTED GOALS CAN GIVE US AN INDICATION OF HOW MANY GOALS A PLAYER OR TEAM SHOULD HAVE SCORED ON AVERAGE, GIVEN THE SHOTS THEY HAVE TAKEN.

I'm not sure why are you getting so aggressive, if you're not interested in this topic and trying to prove that it's all bullshit because somebody scored or not scored a goal against the odds then it's pointless discussion. You seem well oriented in those parameters so not sure why are you pressing me into some definitions. I stated clearly I want to analyze trends for United and see how it correlates to actual score, not individual xG values.
 

Oranges038

Full Member
Joined
Oct 19, 2020
Messages
11,783
Let me quote OPTA for you
EXPECTED GOALS (XG)
EXPECTED GOALS (XG) MEASURES THE QUALITY OF A SHOT BASED ON SEVERAL VARIABLES SUCH AS ASSIST TYPE, SHOT ANGLE AND DISTANCE FROM GOAL, WHETHER IT WAS A HEADED SHOT AND WHETHER IT WAS DEFINED AS A BIG CHANCE.

ADDING UP A PLAYER OR TEAM’S EXPECTED GOALS CAN GIVE US AN INDICATION OF HOW MANY GOALS A PLAYER OR TEAM SHOULD HAVE SCORED ON AVERAGE, GIVEN THE SHOTS THEY HAVE TAKEN.

I'm not sure why are you getting so aggressive, if you're not interested in this topic and trying to prove that it's all bullshit because somebody scored or not scored a goal against the odds then it's pointless discussion. You seem well oriented in those parameters so not sure why are you pressing me into some definitions. I stated clearly I want to analyze trends for United and see how it correlates to actual score, not individual xG values.
Sorry, That wasn't aimed at you and I dont think I am being aggressive at all. I've done a bit of digging since yesterday and I can see the point about the comparisons for overall chances created against goals scored over a period of a season or a managers tenure to date0.

But to be putting weighted numbers on individual chances is not very accurate as there are so many different variables within each chance, no two are ever the exact same. This is why I was asking the other poster to provide the facts for the chance to be labelled as .03 xg. Otherwise it's just as arbitrary as saying he should be scoring from there.
 

Gabagoo

New Member
Newbie
Joined
Oct 15, 2020
Messages
246
The one thing that I think this thread is missing is the comparison to others team's trends.

It's fine to say "we're improving, keep everything the same" but... I'm afraid that's a bit weak, in all honesty.

If Liverpool and City are improving at higher rates then that means that we're improving compared to ourselves but actually getting left further behind the best teams. If Chelsea, Spurs, Arsenal and even Leicester are improving at equal or different rates then we're not going to improve relative to our rivals.
 

Borys

Statistics Wizard
Joined
May 10, 2013
Messages
9,023
Location
Bielsko Biala, Poland
Don't let this discourage you. I'm interested in you future posts about the same thing - data analysis.

I agree with your other points, we are improving... but is it enough? no. Accepting it meant we're accepting low standard. But it's nice to see and know that we are improving this season, even if it's a bit and still need seriously many steps. It's obvious we're improving though, just from seeing the games. The same problems daily seasons of before still exist <-- that's what frustrate people and argued we are just not improving overall.

Anyway.
I just don't get your two... "conclusions"?

First one.

Isn't people moaning about it being a lucky win because.. (1) we didn't concede the penalty fouled by Bruno (for me it's a penalty), and the (2)retaking of our penalties which thankfully Bruno scored in the 2nd. Our finishing is so bad, we were basically heading into a draw.

I also have problems with your statement stating using xg that if we play the game again, it'll be what the xg stated. I think this is the problem with people using data as the deciding point in argument - the assumption that data will (1) still be the same and (2) will decide clearly on average of what "should" happen. It doesn't work that way and I'm sure you'll know this or maybe not yet.

A) If we play the game again, won't the stats be different? The problem is this idea that playing the "same" games again will yield the same xg stats. No it won't. The stats will be different as players performing differently eg. if let's say West Brom won the penalty, it'll spark different reactions to their players as does our players, then creating different xg stats. Every sequence of events per "same" game happen before will affect what'll happen later, thereby changing the whole course of events and the xg stats. so every repetitive games will be different and shall have different unique xg stats.

B) Of course, even if you're still arguing with that line of argument that the xg will maintain being consistent, then why only xg that remains consistent? What happen on the pitch and stats-wise conversion rate should also maintain consistent no? I mean if we'll be creating a lot of that chances again if we repeat the same games, then we will also be wasting those chances again in the same repeated games.

So using xg to maybe prove whatever you're trying to prove there is just full of flaws. Actually it's not quite clear what you're trying to prove here. Is it lucks?

Second one.

Don't see the point here.

Is it to show that we are better than Spurs when facing the mighty West Brom?

It's two different games where West Brom performed differently - their players performed better vs Spurs, but performed worser vs us except for their keeper who had a average game vs Spurs and "worldly" performance vs us. Then again when opposing team did many shit shootings and the ball goes towards you, then it's easier to save and appear great. Credit to Johnstone for stepping-up against us, and their whole team stepped up vs Spurs.

Also our players and Spurs players performed differently and uniquely as they did for that two different games. It's also on two different days. I'm sure you know about form - there's the team form and then there's the individual players form.

So I still don't get what perspective needed changing with this data.
Thank you for your post. I get the idea why this WBA win is considered as "lucky", but I just don't agree with the notion that there is no progress. Anyway, let me explain to you how I understand the "played this game 100 times" sentence. Please be aware that it's all very new to me (I just learned about xG parameters a couple of days ago), so this will be a good mental exercise.

When I said “if this game was played 100 times, we would score 2,44 goals on average” I mean it like a mathematical model. In every calculation you use the same parameters (xG for individual attempts), but it doesn’t mean you will always get the same result.

So lets work on an example and we’ll see if there’s logic in my understanding.

Team A xG score is 0,7, which is a sum of only two attempts:
  1. xGg=0,5 (Gerrard slipped and striker is 1 on 1 with the keeper, expected to score 5/10 such chances)
  2. xGh=0,2 (header in the box from a corner, expected to score 2/10 of those)
Now, each individual xG is either a goal or a miss. So, Team A will score between 0-2 goals (2 attempts) depending on luck and other factors not included in this model (not important because this calculation is done multiple times). The table shows a possible outcome from 10 calculations:

So there is a number of possible outcomes, in 6 scenarios Team A scores a goal, in one scenario scored 2 goals, and in 4 scenarios they don’t score. So, if you sum up the number of scored goals and divide by number of calculations, you get xG for game=0,7. Obviously, the number of calculations will be higher so it’s much simplified explanation.

Now let’s consider Team B, xG is 2,0, and it’s a sum of 6 attempts:
  1. xGp=0,8 (penalty expected to score 8/10 times)
  2. xGh=0,2 (header in the box from a corner, expected to score 2/10 of those)
  3. xGs=0,25 (decent shooting chance, expected to score ¼)
  4. xGs=0,25 (decent shooting chance, expected to score ¼)
  5. xGs=0,25 (decent shooting chance, expected to score ¼)
  6. xGs=0,25 (decent shooting chance, expected to score ¼)

Team B will score between 0 and 6 goals. The number of possible scenarios is much bigger, so let’s try another approach and lets assume:
This table shows that the most likely Team B is to score 2 goals considering the chances quality and their number.

So I’m not saying next time we will beat WBA by 2 goals, I’m saying that a model created based on attempted shots shows we were most likely to win by 2 goals. In some less likely scenarios we score 1 or even 0 goals, but that’s not a reason to panic really. Against WBA we got the highest xG in all our games, which is good. We kept them fairly quiet too with xGA<0,5, what suggests or defensive side of the game is very stable whoever we play (obviously, we were set up defensively against Chelsea and Arsenal but the point stands).

What is interesting is on how many occasions the final score is not what the model would’ve predicted, and why.

Regarding the comments on Spurs vs WBA game, the point was to show that if we struggled against WBA, then so did Tottenham. No direct comparison obviously, but again, hardly a reason to panic.

Again, I don't want to act omniscient, I hope we can discuss ideas here and feel free to correct me.