ManchesterYoda's player ratings

[ManchesterYoda]

These player ratings are based on Opta stats using my own algorithm.
I will update after 30 league games.

Spoiler

Spoiler

Brief explanation of the algorithm:
Players start each match with a rating of 6.5, this increases with a positive contribution such as scoring a goal +0.8 or making an interception +0.1 and decreases with a negative contribution such as missing a big chance -0.2 or losing possession due to bad control -0.05.
If someone plays for 90 mins in one game and gets a rating of 8 and 10 mins in another game and gets a rating of 6.5, their average rating is 8 x 90 + 6.5 x 10 / 100 = 7.85, not 8 + 6.5 / 2 = 7.25.

 

[Giuliano Maiorana]

The fact that young has virtually a 7 means this is basically null and void

 

[Crashoutcassius]

pogba our third best player is bizarre. has he had 2 good games?

 

[King Andow]

How is Young higher than James? You don't count bad tackles and misplaced passes?

 

[Jonno]

Young rated higher than James. /Thread

 

[Wumminator]

Young has had a good year. The hate he gets is bizarre.

 

[ManchesterYoda]

How is Young higher than James? You don't count bad tackles and misplaced passes?

Yes both are counted, for Young, for James, for all players.

 

[Classical Mechanic]

Young has created some big chances this year, from corners specifically. He has our second highest xA after Rashford. Pogba has the highest xA per 90 mins.

 

[Revan]

Your algorithm seem to be a set of ad-hoc crafted rules, which essentially makes it a very bad algorithm.

 

[DWelbz19]

No idea about the actual algorithm itself, but the ranking of players seems about right. Maguire might be getting a bit of a boost because, like WhoScored, the algorithm might be preferential to aerial duels won?

 

[Moved To Differently]

I like what you've done, weighting the performance rating against time played so the absolute rating is a better indicator of who should be playing based on their consistency for longer periods but there's a myriad of complexity definitionally speaking when it comes to the positive/ negative contribution coefficients. Like, how exactly do you define a 'big chance' and is it relative to the particular purpose of the player. We'd all agree a main striker missing a penalty (even against Krool) probably constitutes a big chance missed but what about the ball dropping to a centre back on his weak foot from a corner? Also losing possession - is it relative to the position on the pitch where possession was lost? Is it still possession lost when it was a bold, penetrating pass in the final third that the recipient didn't anticipate?

I think rather than do a multivariate analysis of every players' performance across such a wide range of possible contributions it might be better to do something unidimensional like 'passing' and then you could develop a more meaningful image of each player's contribution.

How are you collecting the data?
Do you have more information on your assigned contribution coefficients?

 

[ManchesterYoda]

No idea about the actual algorithm itself, but the ranking of players seems about right. Maguire might be getting a bit of a boost because, like WhoScored, the algorithm might be preferential to aerial duels won?

Passing and defending is where Maguire's positive contribution mostly comes from. He usually has a positive for both, whereas other defenders might have negative for passing and positive for defending eg Wan-Bissaka.

 

[ManchesterYoda]

I have updated the player ratings after 20 games. If anyone is wondering why a player is not included it is because they have played less than 25%.

 

[lsd]

Maguire being our best player is an absolute joke

 

[432JuanMata]

Peirera is the standout has played over 50% of games and is consistently crap

 

[Classical Mechanic]

@ManchesterYoda

How do you deal with sub apps?

 

[The Cat]

Looks a load of nonsense sorry.

 

[A-man]

@ManchesterYoda

How do you deal with sub apps?

I think it was stated in the explanation:
”If someone plays for 90 mins in one game and gets a rating of 8 and 10 mins in another game and gets a rating of 6.5, their average rating is 8 x 90 + 6.5 x 10 / 100 = 7.85, not 8 + 6.5 / 2 = 7.25.”

 

[Classical Mechanic]

I think it was stated in the explanation:
”If someone plays for 90 mins in one game and gets a rating of 8 and 10 mins in another game and gets a rating of 6.5, their average rating is 8 x 90 + 6.5 x 10 / 100 = 7.85, not 8 + 6.5 / 2 = 7.25.”

Cheers, should have read that!

 

[SER19]

Not sure what people are so put out about. The top 5 players in the categories that have featured significantly have easily been our top 5 performers.

Rashford, maguire, wan bissaka, mctominay, Fred

Lingard, Shaw, pereira the weaker performers also accurate

 

[arthurka]

Young has had a good year. The hate he gets is bizarre.

Agreed he has been a very solid support player for us. But I am very glad he isn't starting games.

 

[Gandalf Greyhame]

Players start each match with a rating of 6.5, this increases with a positive contribution such as scoring a goal +0.8 or making an interception +0.1 and decreases with a negative contribution such as missing a big chance -0.2 or losing possession due to bad control -0.05.
If someone plays for 90 mins in one game and gets a rating of 8 and 10 mins in another game and gets a rating of 6.5, their average rating is 8 x 90 + 6.5 x 10 / 100 = 7.85, not 8 + 6.5 / 2 = 7.25.[/spoiler]

I like the time-weighted ratings, and bucketing of players by aggregated time (which weeds out a solitary 9/10 against a season long 8/10), but I have some suggestions:

1) Why is 6.5 the default? If your scale is from 0-10, doesn't 5 make statistical sense?
2) What are the complete set of contributions? Where are you getting them from?
- Goals
- Interceptions
- Missing a big chance
- Bad Control
- ?
3) On what basis have you decided that a scoring a goals is four times good and missing a big chance is one time bad?
4) Can you make the ratings subject to how the team's metrics look? (Missing a big chance is less bad for City than for Leicester)
5) Can you make it specific per position? (De Gea making an error is always a goal, Martial losing the ball not so much)

I like things like these, if you want I can help with the coding as well (I do well with Python and data)

 

[Gandalf Greyhame]

Your algorithm seem to be a set of ad-hoc crafted rules, which essentially makes it a very bad algorithm.

This is a concern, yep. What would you change?

 

[Scarecrow]

A rare interesting thread in the football forum. Good work @ManchesterYoda. Please keep it up.

 

[Member 119614]

So many players who get bizarre hate in our squad. The redcafe is a collective that becomes 1 blob, an organism of negativity. It consumes every new member almost immediately. The entire forum hates the likes of Mata, Jones, Young and the nr 1 villain Lingard. Fred used to be among them too but miraculously he became one of the good guys 2 months ago so there's hope.

I'm the only person here that's not yet been devoured by the predatory beast that is the red blob. I must be the special one. Worship me.

 

[jem]

Young has had a good year. The hate he gets is bizarre.

Yeah I was thinking about this the other day. I think Young has been very solid this year. Granted, he was shite at the end of last season, but for the most part, I think he's done a solid job in the fullback position - the definition of valuable squad depth, in my opinion.

 

[11101]

I like the weighting but the underlying data looks way off. Some very wrong ratings for certain players.

 

[sunama]

There is one common stat in all those ratings.
Lingard is consistently our worst player.

 

[sunama]

I'm the only person here that's not yet been devoured by the predatory beast that is the red blob. I must be the special one. Worship me.

You on your period?

 

[harms]

1) Why is 6.5 the default? If your scale is from 0-10, doesn't 5 make statistical sense?

If a player had a 6/10 game, it was not a good game. A positive rating usually starts from 6,5 or 7/10

 

[Ekeke]

Best to split between position. If a player is filling in with a position that isnt really what hes there for, its going to bomb their rating that they have a little bit of versatility and took one for the team.

 

[Amerifan]

A rare interesting thread in the football forum. Good work @ManchesterYoda. Please keep it up.

Agree. More of this please. Much more compelling and thought provoking than the “I hate something/everything” threads that are drowning the forum.

 

[NewGlory]

Validity of a model like this is all about choosing the coefficients correctly (the +0.8, +0.1, -0.2, -0.05. etc.). Your choice of the numbers in your model - are they based on anything except for intuition? And have you used any approach of validating different numbers to find optimal versions?

 

[Scarecrow]

Validity of a model like this is all about choosing the coefficients correctly (the +0.8, +0.1, -0.2, -0.05. etc.). Your choice of the numbers in your model - are they based on anything except for intuition? And have you used any approach of validating different numbers to find optimal versions?

It's a good point but what else can they be based on, if not intuition? Can't think of anything, myself.

 

[NewGlory]

It's a good point but what else can they be based on, if not intuition? Can't think of anything, myself.

I am not a data scientist but I assume you could run a Monte Carlo simulation, or any other optimizer over a training/historical set of data to find better-fitting values for the coefficients than what pure intuition delivers in this example

 

[ManchesterYoda]

I like what you've done, weighting the performance rating against time played so the absolute rating is a better indicator of who should be playing based on their consistency for longer periods but there's a myriad of complexity definitionally speaking when it comes to the positive/ negative contribution coefficients. Like, how exactly do you define a 'big chance' and is it relative to the particular purpose of the player. We'd all agree a main striker missing a penalty (even against Krool) probably constitutes a big chance missed but what about the ball dropping to a centre back on his weak foot from a corner? Also losing possession - is it relative to the position on the pitch where possession was lost? Is it still possession lost when it was a bold, penetrating pass in the final third that the recipient didn't anticipate?

I think rather than do a multivariate analysis of every players' performance across such a wide range of possible contributions it might be better to do something unidimensional like 'passing' and then you could develop a more meaningful image of each player's contribution.

How are you collecting the data?
Do you have more information on your assigned contribution coefficients?

I collect the data from Whoscored. Each player has a google spreadsheet. I go through in alphabetical order. I start off on commentary to see minutes played. Then I go to summary match centre tab, look at key passes, assists, inaccurate passes and aerials. Then I go to player stats and note accurate long passes, accurate throughballs and crosses. Then I go to chalkboard and look at shooting, dribbling, possession, tackling, interceptions, clearances, blocks, fouls and errors. Finally if a GK I go through commentary making a note of all saves and goals conceded if any.

 

[ManchesterYoda]

I like the time-weighted ratings, and bucketing of players by aggregated time (which weeds out a solitary 9/10 against a season long 8/10), but I have some suggestions:

1) Why is 6.5 the default? If your scale is from 0-10, doesn't 5 make statistical sense?
2) What are the complete set of contributions? Where are you getting them from?
- Goals
- Interceptions
- Missing a big chance
- Bad Control
- ?
3) On what basis have you decided that a scoring a goals is four times good and missing a big chance is one time bad?
4) Can you make the ratings subject to how the team's metrics look? (Missing a big chance is less bad for City than for Leicester)
5) Can you make it specific per position? (De Gea making an error is always a goal, Martial losing the ball not so much)

I like things like these, if you want I can help with the coding as well (I do well with Python and data)

6.5 is the default mainly because subs are unfairly rated with a 5.x if they only have a few minutes on the pitch. Players only get a 5.x rating if they have had an awful game using my algorithm.

 

[ManchesterYoda]

I am posting the player ratings for the match against Wolves along with Whoscored player ratings for comparison.

7.600 Romero
7.550 Maguire
7.075 Lindelof
7.025 Mata
6.900 Greenwood
6.500 ------------------------------ For players above this line, the positive contribution outweighed the negative contribution and vice versa.
6.500 Chong
6.375 Matic
6.350 Williams
6.300 James
6.025 Young
5.850 Pereira


Whoscored

7.81 Maguire
7.22 Lindelof
6.91 Romero
6.91 Greenwood
6.71 Matic
6.62 Pereira
6.59 Mata
6.51 Williams
6.43 Young
6.30 Chong
6.26 James
6.00 ------------------------------ For players above this line, the positive contribution outweighed the negative contribution and vice versa. Really?


This is one of several reasons I believe my algorithm to be more accurate than Whoscored's algorithm.

 

[TwoSheds]

Well Greenwood was shite so can we just say both your algorithms are bollocks and be done with it?