Dear all, even though I watch the forum, I rarely post here.
I would like with this post to share one more approach on how to obtain player ratings. These have not been thoroughly tested and I mainly post them to obtain your collective wisdom as well.
The main aim in coming up with these ratings is that even though we know that the player roles are mainly cosmetic I still want to have some ratings to compare players in the same position but with different roles. After all, whenever we have an Advanced Forward and a Pressing Forward in our formation, I am sure that we all choose who to place where according to their attributes.
So far we have Mark's ratings and ykykyk ratings which are not based on player roles. We also have Squirrel_Plays and FMDataLab's ratings which I find heavily dependent on the player role and therefore I do not trust that much. I do appreciate their efforts a lot though and use their tools when playing FM. I also used them to experiment with the new ratings.
So here is my approach:
Just like Mark's ratings I found it convenient the sum of the weights that I place to the attributes to be (roughly) equal to 500. After some experimentation I found out that I want the Meta Attributes to account for 75 from these 500 points. (So they weight 15% of a player's rating.) Before complaining that this is too low please note that these attributes will get an extra weighing later as well. I initially gave them a weight of 50% but then I found out that the players' ratings mainly depended on these attributes irrespective of their position. (E.g. a striker with high pace acceleration gets very high ratings across all positions.)
I also decided (without too much testing) for 250 out of these 500 points to come from FM Scout's positional weight table. (Note that Meta Attributes also get weights from here.)
Finally the other 175 points will come from FM's positional roles.
Even though I feel that the 75 out of 500 points for Meta Attributes seems reasonable (see the final weight later), maybe the 175 points for the FM positional roles is too much and needs to be reduced.
So here is how I distribute these weights. My example will be Acceleration for a Striker with an Attacking Forward role.
For Meta Attributes, I start with the Player Attribute Testing results for FM24. I first give to each attribute a value equal to the number of points exceeding the points achieved by the Default Attributes when increased by +5. Due to RNG issues some attributes happen to get a negative value but we ignore that and make it equal to 0.
Example: The default attributes score 60.0 points while Acceleration + 5 scores 80.4 points, so Acceleration gets a value of +20.4.
The total sum of all these values turns out to be 95.3. Since Acceleration gets a 20.4 out of 95.3, then it gets 20.4 * 75/95.3 = 16.05 out of 75 from the Meta Attributes Weight.
Acceleration is also a 10 in the FM Scout's positional weight for a Striker. The total sum of all weights for a striker is 123.5 so acceleration gets a weight of 10 * 250 / 123.5 = 20.24 for the Striker position.
Finally for the positional roles, Acceleration is very important for the Advanced Forward position, so initially I give it a value of 5. I give a value of 3 to important attributes and a value of 1 to other attributes. This is done similarly by Squirrel_Plays I think but I don't have a strong reasoning why I give these weights. I want to give 1 to non-important attributes and have a linear increase for the important and very important attributes. Weights of 1,2,3 don't seem correct because surely an important attribute is more than twice as important as corner taking for example. So I settled for 1,3,5.
The total sum of weights for the AF role is 89 so I give acceleration 5*175/81 = 10.8 out of 175 for the Role weight.
In total the acceleration weight for the position of ST and the role of AF is 16.05 + 20.24 + 10.8 = 47 when rounded down.
You can find my weights using this approach for some positions / roles in the screenshot below. I have not computed weights for more roles because I am sure that based on your input I may have to reconsider parts of my approach. So I would really like to hear your thoughts on this. Note that for most attributes the weights lie between the weights of Mark and ykykyk (or are not far off from them) when arranged to have a total weight sum of 500. So at least I am somewhat reassured that my approach is not too silly. There are some exceptions though. A notable one is that I tend to have lower weight on stamina across most positions/roles. I am not sure if this is due to the fact that stamina doesn't play too much a role in the Player Attribute Testing (e.g. because all players start with 100% condition) but plays a more significant role otherwise.
Well, interesting for sure. I have two questions that I would like to ask you, if I may.
1) I see that you take the values for free kicks, corners, penalties, and long throws into your calculation. Why is that? I mean, if a player is not good at set pieces, it doesn't mean he could not be a world-class defender, midfielder, or striker, right? Even for goalkeepers, you include these values in your calculations. There is no way you're going to let your goalkeeper take your set pieces... I hope, at least.
2) Do I understand correctly that you're trying to rebuild the CA calculation of FM itself or trying to get as close as possible to the game’s calculation? Can I ask why you are doing that (no offense here I tried the same)? Is it because you want to compare the outcome of your calculation with the actual CA of the game, to see if the calculation is correct?
Because FM uses a little 'trick' to calculate the CA and PA of players in the game. I've also heard about the 500, but that is not 100% accurate, I believe (especially when players can play in more than one position). By trying to recreate the game's calculation (if that is what you're trying), I believe you're making it harder to calculate the correct score, and the chance of making a mistake is increased too.
Note that the game also gives a higher score to players who can play in more than one position. So a player who can only play as a striker will have a lower CA than an other player with the exact same values of attributes who can also play as an attacking midfielder and winger. Because of this, I stopped trying to calculate a CA based on the way the game does. It wouldn't make any sense in my point of view and therefore, I took a different approach.
If you are not trying to recalculate the game method but want something similar, you should check out "Evidence Based Football Manager" on YouTube (perhaps you already did). He tried to rebuild the game's calculation. Perhaps you will find something of interest that you can use in your method. The video title is "Relationship between Attributes and CA in Football Manager." Also, the pre-game editor will show you the weighting per attribute for the CA calculation. It might be interesting to compare your weighting to it and analyze the differences.
Hopes this helps you a bit in your player rating method.
Before replying to the comments, let me mention that I am not yet happy with my ratings. If we look at the post here, when we change the players to 20 for Meta attributes and 7 for non-Meta attributes their ratings are worse but they actually perform better. (The same happens with the MDW22 and the ykykyk ratings.)
Their ratings need to be better, not worse, but how much better? About as much as if we do +5 to each player's dribbling or +5 to each player's jumping reach.
I am in the process of fixing this but I am not yet done. My main issue at the moment is that my updated (not yet shared) weights fixing this issue have other problems. The major one is that they severely underestimate DMs. So much so that if I pick a DM with CA 140-150 at random, his rating is worse than the DM-rating of a random non-DM with CA 140-150.
I don't yet have any clue on what to do with this. I tried to look at each individual attribute and noticed the following interesting feature:
Group A = All players with CA from 140 to 150 who cannot play at the DM position. Group B = All players with CA from 140 to 150 who are natural at DM and cannot play at any of DC, DR, DL, AMR, AML, ST. (I did not remove the MC position because then no players remain.)
Average Acceleration in Group A: 14 Average Acceleration in Group B: 13
Average Pace in Group A: 14 Average Pace in Group B: 13
Average Jumping Reach in Group A: 11.5 Average Jumping Reach in Group B: 10.5
In every other attribute the Group B average is better or very close to the Group A average. (The Dribbling averages are 12 for both groups.)
This is what seems to cause non-DM players to have in general better DM ratings with my weights than DM players of the same CA. I will need to think what to do with this issue. Perhaps nothing should be done but I cannot test it.
joshua said: could you post the spreadsheet file? Expand
Sure but as I said I am not yet happy with them. Once I have something better I will share the excel file for your convenience.
Robby said: Well, interesting for sure. I have two questions that I would like to ask you, if I may.
1) I see that you take the values for free kicks, corners, penalties, and long throws into your calculation. Why is that? I mean, if a player is not good at set pieces, it doesn't mean he could not be a world-class defender, midfielder, or striker, right? Even for goalkeepers, you include these values in your calculations. There is no way you're going to let your goalkeeper take your set pieces... I hope, at least. Expand
I include them but each one weights about 4 out of 500, i.e. roughly 0.8%. So overall, if you have a player with 70% rating with 10 free kicks and another one with the same attributes but 20 free kicks, he would have a 70.4% rating. Is the difference that much? These are also taken into account by MDW22 and ykykyk ratings. (But not the ykykyk minus 50.) In my new ratings that I did not yet share these weights are further reduced to about 2 out of 500.
By the way I don't have goalkeeper ratings yet. However why not let Chilavert take the free kick?
Robby said: 2) Do I understand correctly that you're trying to rebuild the CA calculation of FM itself or trying to get as close as possible to the game’s calculation? Can I ask why you are doing that (no offense here I tried the same)? Is it because you want to compare the outcome of your calculation with the actual CA of the game, to see if the calculation is correct?
Because FM uses a little 'trick' to calculate the CA and PA of players in the game. I've also heard about the 500, but that is not 100% accurate, I believe (especially when players can play in more than one position). By trying to recreate the game's calculation (if that is what you're trying), I believe you're making it harder to calculate the correct score, and the chance of making a mistake is increased too.
Note that the game also gives a higher score to players who can play in more than one position. So a player who can only play as a striker will have a lower CA than an other player with the exact same values of attributes who can also play as an attacking midfielder and winger. Because of this, I stopped trying to calculate a CA based on the way the game does. It wouldn't make any sense in my point of view and therefore, I took a different approach.
If you are not trying to recalculate the game method but want something similar, you should check out "Evidence Based Football Manager" on YouTube (perhaps you already did). He tried to rebuild the game's calculation. Perhaps you will find something of interest that you can use in your method. The video title is "Relationship between Attributes and CA in Football Manager." Also, the pre-game editor will show you the weighting per attribute for the CA calculation. It might be interesting to compare your weighting to it and analyze the differences.
Hopes this helps you a bit in your player rating method. Expand
Yes and no. Ignoring a player's multiple positions, in my calculations the CA accounted for 50% of the weight. I took this directly from the FM Scout's weights. I feel that this is necessary (maybe with a different weight) because whatever "Evidence Based Importance of Attributes" I have seen so far (maybe I missed something) is Importance of Attributes for the team overall rather than per position. The only way I know to account for position is to take into consideration the CA weights.
My main task/hope is to give a single number (rating) to each player for each position (Mark's weights do an excellent job at that) but go a step further and give a rating per role as well. I know that this is mostly insignificant but as I mentioned we all distinguish between an Advanced and a Pressing Forward based on the attributes. I want to do it based on this single number.
Dear all, even though I watch the forum, I rarely post here.
I would like with this post to share one more approach on how to obtain player ratings. These have not been thoroughly tested and I mainly post them to obtain your collective wisdom as well.
The main aim in coming up with these ratings is that even though we know that the player roles are mainly cosmetic I still want to have some ratings to compare players in the same position but with different roles. After all, whenever we have an Advanced Forward and a Pressing Forward in our formation, I am sure that we all choose who to place where according to their attributes.
So far we have Mark's ratings and ykykyk ratings which are not based on player roles. We also have Squirrel_Plays and FMDataLab's ratings which I find heavily dependent on the player role and therefore I do not trust that much. I do appreciate their efforts a lot though and use their tools when playing FM. I also used them to experiment with the new ratings.
So here is my approach:
Just like Mark's ratings I found it convenient the sum of the weights that I place to the attributes to be (roughly) equal to 500. After some experimentation I found out that I want the Meta Attributes to account for 75 from these 500 points. (So they weight 15% of a player's rating.) Before complaining that this is too low please note that these attributes will get an extra weighing later as well. I initially gave them a weight of 50% but then I found out that the players' ratings mainly depended on these attributes irrespective of their position. (E.g. a striker with high pace acceleration gets very high ratings across all positions.)
I also decided (without too much testing) for 250 out of these 500 points to come from FM Scout's positional weight table. (Note that Meta Attributes also get weights from here.)
Finally the other 175 points will come from FM's positional roles.
Even though I feel that the 75 out of 500 points for Meta Attributes seems reasonable (see the final weight later), maybe the 175 points for the FM positional roles is too much and needs to be reduced.
So here is how I distribute these weights. My example will be Acceleration for a Striker with an Attacking Forward role.
For Meta Attributes, I start with the Player Attribute Testing results for FM24. I first give to each attribute a value equal to the number of points exceeding the points achieved by the Default Attributes when increased by +5. Due to RNG issues some attributes happen to get a negative value but we ignore that and make it equal to 0.
Example: The default attributes score 60.0 points while Acceleration + 5 scores 80.4 points, so Acceleration gets a value of +20.4.
The total sum of all these values turns out to be 95.3. Since Acceleration gets a 20.4 out of 95.3, then it gets 20.4 * 75/95.3 = 16.05 out of 75 from the Meta Attributes Weight.
Acceleration is also a 10 in the FM Scout's positional weight for a Striker. The total sum of all weights for a striker is 123.5 so acceleration gets a weight of 10 * 250 / 123.5 = 20.24 for the Striker position.
Finally for the positional roles, Acceleration is very important for the Advanced Forward position, so initially I give it a value of 5. I give a value of 3 to important attributes and a value of 1 to other attributes. This is done similarly by Squirrel_Plays I think but I don't have a strong reasoning why I give these weights. I want to give 1 to non-important attributes and have a linear increase for the important and very important attributes. Weights of 1,2,3 don't seem correct because surely an important attribute is more than twice as important as corner taking for example. So I settled for 1,3,5.
The total sum of weights for the AF role is 89 so I give acceleration 5*175/81 = 10.8 out of 175 for the Role weight.
In total the acceleration weight for the position of ST and the role of AF is 16.05 + 20.24 + 10.8 = 47 when rounded down.
You can find my weights using this approach for some positions / roles in the screenshot below. I have not computed weights for more roles because I am sure that based on your input I may have to reconsider parts of my approach. So I would really like to hear your thoughts on this. Note that for most attributes the weights lie between the weights of Mark and ykykyk (or are not far off from them) when arranged to have a total weight sum of 500. So at least I am somewhat reassured that my approach is not too silly. There are some exceptions though. A notable one is that I tend to have lower weight on stamina across most positions/roles. I am not sure if this is due to the fact that stamina doesn't play too much a role in the Player Attribute Testing (e.g. because all players start with 100% condition) but plays a more significant role otherwise.
could you post the spreadsheet file?
Well, interesting for sure. I have two questions that I would like to ask you, if I may.
1) I see that you take the values for free kicks, corners, penalties, and long throws into your calculation. Why is that? I mean, if a player is not good at set pieces, it doesn't mean he could not be a world-class defender, midfielder, or striker, right? Even for goalkeepers, you include these values in your calculations. There is no way you're going to let your goalkeeper take your set pieces... I hope, at least.
2) Do I understand correctly that you're trying to rebuild the CA calculation of FM itself or trying to get as close as possible to the game’s calculation? Can I ask why you are doing that (no offense here I tried the same)? Is it because you want to compare the outcome of your calculation with the actual CA of the game, to see if the calculation is correct?
Because FM uses a little 'trick' to calculate the CA and PA of players in the game. I've also heard about the 500, but that is not 100% accurate, I believe (especially when players can play in more than one position). By trying to recreate the game's calculation (if that is what you're trying), I believe you're making it harder to calculate the correct score, and the chance of making a mistake is increased too.
Note that the game also gives a higher score to players who can play in more than one position. So a player who can only play as a striker will have a lower CA than an other player with the exact same values of attributes who can also play as an attacking midfielder and winger. Because of this, I stopped trying to calculate a CA based on the way the game does. It wouldn't make any sense in my point of view and therefore, I took a different approach.
If you are not trying to recalculate the game method but want something similar, you should check out "Evidence Based Football Manager" on YouTube (perhaps you already did). He tried to rebuild the game's calculation. Perhaps you will find something of interest that you can use in your method. The video title is "Relationship between Attributes and CA in Football Manager." Also, the pre-game editor will show you the weighting per attribute for the CA calculation. It might be interesting to compare your weighting to it and analyze the differences.
Hopes this helps you a bit in your player rating method.
Before replying to the comments, let me mention that I am not yet happy with my ratings. If we look at the post here, when we change the players to 20 for Meta attributes and 7 for non-Meta attributes their ratings are worse but they actually perform better. (The same happens with the MDW22 and the ykykyk ratings.)
Their ratings need to be better, not worse, but how much better? About as much as if we do +5 to each player's dribbling or +5 to each player's jumping reach.
I am in the process of fixing this but I am not yet done. My main issue at the moment is that my updated (not yet shared) weights fixing this issue have other problems. The major one is that they severely underestimate DMs. So much so that if I pick a DM with CA 140-150 at random, his rating is worse than the DM-rating of a random non-DM with CA 140-150.
I don't yet have any clue on what to do with this. I tried to look at each individual attribute and noticed the following interesting feature:
Group A = All players with CA from 140 to 150 who cannot play at the DM position.
Group B = All players with CA from 140 to 150 who are natural at DM and cannot play at any of DC, DR, DL, AMR, AML, ST. (I did not remove the MC position because then no players remain.)
Average Acceleration in Group A: 14
Average Acceleration in Group B: 13
Average Pace in Group A: 14
Average Pace in Group B: 13
Average Jumping Reach in Group A: 11.5
Average Jumping Reach in Group B: 10.5
In every other attribute the Group B average is better or very close to the Group A average. (The Dribbling averages are 12 for both groups.)
This is what seems to cause non-DM players to have in general better DM ratings with my weights than DM players of the same CA. I will need to think what to do with this issue. Perhaps nothing should be done but I cannot test it.
joshua said: could you post the spreadsheet file?
Sure but as I said I am not yet happy with them. Once I have something better I will share the excel file for your convenience.
Robby said: Well, interesting for sure. I have two questions that I would like to ask you, if I may.
1) I see that you take the values for free kicks, corners, penalties, and long throws into your calculation. Why is that? I mean, if a player is not good at set pieces, it doesn't mean he could not be a world-class defender, midfielder, or striker, right? Even for goalkeepers, you include these values in your calculations. There is no way you're going to let your goalkeeper take your set pieces... I hope, at least.
I include them but each one weights about 4 out of 500, i.e. roughly 0.8%. So overall, if you have a player with 70% rating with 10 free kicks and another one with the same attributes but 20 free kicks, he would have a 70.4% rating. Is the difference that much? These are also taken into account by MDW22 and ykykyk ratings. (But not the ykykyk minus 50.) In my new ratings that I did not yet share these weights are further reduced to about 2 out of 500.
By the way I don't have goalkeeper ratings yet. However why not let Chilavert take the free kick?
Robby said: 2) Do I understand correctly that you're trying to rebuild the CA calculation of FM itself or trying to get as close as possible to the game’s calculation? Can I ask why you are doing that (no offense here I tried the same)? Is it because you want to compare the outcome of your calculation with the actual CA of the game, to see if the calculation is correct?
Because FM uses a little 'trick' to calculate the CA and PA of players in the game. I've also heard about the 500, but that is not 100% accurate, I believe (especially when players can play in more than one position). By trying to recreate the game's calculation (if that is what you're trying), I believe you're making it harder to calculate the correct score, and the chance of making a mistake is increased too.
Note that the game also gives a higher score to players who can play in more than one position. So a player who can only play as a striker will have a lower CA than an other player with the exact same values of attributes who can also play as an attacking midfielder and winger. Because of this, I stopped trying to calculate a CA based on the way the game does. It wouldn't make any sense in my point of view and therefore, I took a different approach.
If you are not trying to recalculate the game method but want something similar, you should check out "Evidence Based Football Manager" on YouTube (perhaps you already did). He tried to rebuild the game's calculation. Perhaps you will find something of interest that you can use in your method. The video title is "Relationship between Attributes and CA in Football Manager." Also, the pre-game editor will show you the weighting per attribute for the CA calculation. It might be interesting to compare your weighting to it and analyze the differences.
Hopes this helps you a bit in your player rating method.
Yes and no. Ignoring a player's multiple positions, in my calculations the CA accounted for 50% of the weight. I took this directly from the FM Scout's weights. I feel that this is necessary (maybe with a different weight) because whatever "Evidence Based Importance of Attributes" I have seen so far (maybe I missed something) is Importance of Attributes for the team overall rather than per position. The only way I know to account for position is to take into consideration the CA weights.
My main task/hope is to give a single number (rating) to each player for each position (Mark's weights do an excellent job at that) but go a step further and give a rating per role as well. I know that this is mostly insignificant but as I mentioned we all distinguish between an Advanced and a Pressing Forward based on the attributes. I want to do it based on this single number.