OVERALL RATINGS FORMULA

[G-Force]

Hi,

I'm looking for the formula that EA uses to calculate the OVERALL ratings in NBA Live 2005.

It's not easy, but it's finding the algorhytm would be great!

All comments are more than welcome.

Thanks,
G-Force/EuroSlam

[birovr]

Hi there,
just seen your post...

I tried to calculate the algorithm as well, and posted in a other thread.
My simulation is:

99-(5400-AVERAGE(all ratings)*70,285)/40

I dunno how I've got to these values... played in excel and found an approssimation that is quite working.

I think the formula is far more complex (reckon the average of all ratings is not just a plain average but uses some weights depending on the position - big guys rebounds count more, guards FG counts more), but it loooks a good approssimation since is max 2/3 points different from the calculated one

For the ratings use all 20 major ratings that are in the game (not scorearea for example).

I'm also trying to figure out an alghoritm/formula to automatically calculate ratings starting from original stats, so that you can easilly update players during the season by importing stats or calculate historically rosters (of course some ratings cant be calculated - strenght/jump/quick/speed... but the major would be). Good thing with this formula is that there would be a standard for all player (i.e if Chamberlain is 99 in rebounding all other players would be scaled to his rating)

Have you ever worked on this topic? Anybody else?

If you are interested in this topic, contact me here or at birovr@libero.it

Cheers,
B

[HonorGlow]

I'd actually be interested in working with you on this birovr as I think it'd be a handy tool to provide at liveGMC.com but it'd be awefully awefully tough we'd have to base a lot on overall tendencies because specs like strength/jump/quick/speed can't be easily determined by mere stats... PM and let's get the ball rolling!

[ParoXsitiC]

I believe the best way to get a good formula similar to how EA's calculates the overall rating is by using all the infomation and subject it to multiple linear regression.

Most likey I would use different regression formulas based on the players poistion because it would seem like each player's overall rating would be based on their objective in the game and how well they accomplish it.


I have made an excel thing which takes linear weights to the specific ratings you'd liekt he overall to be based on. Any customization you need to help you I am willing to do just contact me

http://nprt.byethost11.com

[HonorGlow]

Is there any way you can post the linear weights formulas for each position here, ParoXsitiC?

I checked out your 'download' site and that looks like a very awesome tool. I'd like to develop a web based version of the tool for everyone to use and be happy and gleeful with ...for the few of us who aren't so Excel savvy and some of us...(such as myself) just don't have Excel at all. Not to take anything away from your awesome tool...just trying to help out the community like you mentioned.

Question number 2: Do any of you guys know a good defensive formula using linear weights on blocks, defensive rebounds, steals etc to obtain a numeric value to show how good a player is defensively? I've been scouring Google but just cant find anything... and I'm pretty sure it's just not as simple as:

Defensive Value = (Blocks + Defensive Rebounds + Steals)/ games played...

I'm trying to extend the Pstat evaluation tool on my site to let a person know which player has the best Assist to Turnover ratio, Defensive man, etc…. Thanks guys!

[ParoXsitiC]

well i tested out some multiple linear regression.

Code: Select all

Term                   Coefficient

Intercept               -36.3815
FG                        0.0881
3P                        0.0760
FT                        0.0831
Inside                    0.0803
Dunk                      0.0741
Dribble                   0.0708
Pass                      0.1010
Steal                     0.0831
Block                     0.1054
Reb-O                     0.0883
Reb-D                     0.0905
Aware-O                   0.0960
Aware-D                   0.0834
Speed                     0.1046
Quick                     0.0917
Jump                      0.0748
Strength                  0.1004
Injury                    0.0835
Fatigue                   0.0909
Primacy                   0.0906


That is for all the players minus the special teams such as the lengends, refs and rookies, etc.

the way its written is
y = a + x1b + x2c + x3d
where a = the y interecept
b = term and x next to the term is its coeffcient.

so to start off...

overall = -36.3815 + (FG * .0881) + (3P * .0760) + (FT * .0831) ..etc

===============================
I am sure EA used linear weights with a semi complex formula and that is why there is so many real numbers and not whole.

here is a following on how some players came up with this formula:

Code: Select all

Last           First         NEW         OLD

Garnett        Kevin          98          98
Bryant         Kobe           94          94
O'Neal         Shaquille      92          92
Duncan         Tim            92          91
McGrady        Tracy          90          90
Pierce         Paul           89          89
Carter         Vince          88          88
Davis          Baron          87          88
Allen          Ray            85          86
Iverson        Allen          84          85
Francis        Steve          84          85
Marbury        Stephon        84          85
Kidd           Jason          85          85
Nowitzki       Dirk           86          85
James          LeBron         85          84
O'Neal         Jermaine       84          84
Kirilenko      Andrei         86          83
Marion         Shawn          84          82
Gasol          Pau            83          82

The new column was generated with the multiple linear regression.
It is not perfect, but im sure you'd have to take the whole numbers and find the exact formula EA used.
If you are interested in how to get mutli linear regression VIA excel check out a program http://www.analyse-it.com/start.htm

[ParoXsitiC]

Block
Speed
Pass
Strength
Aware - O
Quick
Fatigue
Primacy
Reb - D
Reb - O
FG
Injury
Aware - D
FT
Steal
Inside
3P
Jump
Dunk
Dribble


That is the list of important features, the top being the most important based on the regression

[HonorGlow]

Awesome! Thanks! Now, what are your thoughts on finding the best defensive player on your team. Do you think those co-efficients and can 'crossover" to a formula such as:

Defensive value =
[0.0831(# of steals) + 0.1054(# of blocks) + 0.0905( #Reb-D)]
---------------------------------------------------------------------------
(# of total games played)

With these three defensive stats, they're pretty much a 1 to 1 to 1 ratio anyway… but as I know that blocks tend to have more weight in determining MVP(believe me, I have O'neal to evaluate this notion from), it seems that it makes logical sense that your coefficients are pretty safe to draw from IF you were only able to draw from the information from the player stat text file.... What do you think?

[ParoXsitiC]

Well for the start there is no defensive values stored in EA live 2005 persay.
However you can compute them with your own logic.

(%) = how much you think they matter.




(%) Stealing
(%) Blocking
(%) D-Rebounds
------------------------
Fabricated Skills

(Fabricated meaning they are skills that are earned from knowledge of the game and through training.)



Fabricated skills*Defensive Awareness rating
-------------------------
Primary Skills

(This will give you an outcome of their primary D skills, skills they have tweaked and tried to perfect over time, its dependent on their awareness or ability to execute. A person can block is no good if he doesnt have the awareness or ability to know when to block.)


(%) Speed
(%) Quickness
(%) Jump
(%) Strength
--------------------------
Secondary skills

(These are generally skills you were born with. Its almost impossible for shaq to be quick because he was born to be slow...etc)


Primary skills + Secondary skills
--------------------------
Final defensive output.

[ParoXsitiC]

Just as an example.
Lets say that I choose this for Centers:

35 goes to stealing. Centers are not people to steal, but if they are a good stealer than they have something special.
55 goes to rebounding. Its kind of expected for the bigger men to get the rebounds, its not impressive but it sure helps your team out when you get rebounds.
15 for blocks. Blocks are also expected within big men, however I rated them low because they dont always end up in a turn over as do steals and rebounds.

(35) Stealing
(15) Blocking
(55) Rebounds
--------------------
Fabricated Skill = (StealRating*.35) + (BlockRating*.15) + (D-ReboundRating.55)


Speed is 10. Centers normally dont need to run fast.
Quickness 20. Quickness is good. Get the ball, quick turn around and lay it in.
Jump is 30. I shouldnt have to explain.
Strength is 40. Strength is needed for big men as they have to fight off other big men.

========================================

These are just numbers I took off my head and how I would rate a defensive player ( i havent put these numbers in effect to see the outcome). Normally one persons way isnt the same as everyones.

If you are looking for general ways to judge defensive mean based on their stats then check out www.82games.com or www.realgm.com also google the words "NBA player ranking" "NBA player ratings" You should get some good matches. Linear weights in the NBA are kind of old now, they have better ways in ranking.

[ParoXsitiC]

I also want to state that the 1-99 Rating is a good way to get an idea of ones skill, however it is more accurate just to have limitless boundaries. For example
saying Kobe bryant is a 135 in overall. Kevin garnet is 140 and Micheal jordan was a 175. This way a player who is better than MJ can exist and you wouldnt have to downrate everyone else to make up for him.

In the game Larry Bird is rated a 99 in shooting and the like.
Larry Bird has been the best. But its not impossible nor unlikey they there can be better.
If I was to be making up a rating system I would base it on 0-1000.
Zero being that the player has missed every shot, doesnt not get rebounds. Does nothing at all. 1000 being that they never miss a shot, get shitloads of rebounds and the like. Of course this would call for a rating system than will never be breached. Noone will ever score 0 nor 1000, but there is enough numbers to correctly rank people. (unlike in the game were like 3 people tie in 99 in rebounds, however one of them is better im sure)

[ParoXsitiC]

PS: if you havent noticed I dont read the things I type. A lot of spelling and grammar mistakes so please disreguard

[HonorGlow]

Well for the start there is no defensive values stored in EA live 2005 persay.

I think we're mixing things up here… when I meant defensive values, I'm referring to simply how much value a player has as a defensive player… the formula I'm attempting to derive is based off of the mere pstat file that one obtains via the Save stats options….and these are defenitely stored.

…and again I've googled and haven't found anything definitive…based on the 'common stats', 82games.com apparently obtains more obscure stats then the average fan is interested in obtaining AND their formulas are usually derived from them…so that brings us back to square one.

(35) Stealing
(15) Blocking
(55) Rebounds
--------------------
Fabricated Skill = (StealRating*.35) + (BlockRating*.15) + (D-ReboundRating.55)

Steal 0.0831
Block 0.1054
Reb-O 0.0883
Reb-D 0.0905

I know that your Fabricated skill coefficiants are somewhat arbitrary, but should they be somewhat in line with the coefficiants for your linear regression formula?

In terms of mere 'EA value' I know from personal experience that EA values the block vs perhaps the Defensive rebound(which is consistant to your coefficients) as I have Shaq on my team leading man for MVP and all his stats are lower than the other MVP candidates except for his block shots which thrusts him to the front of the pack.

Linear weights in the NBA are kind of old now, they have better ways in ranking

True, NBA.com still uses it so there's still some merit to it. And unfortunately again, the standards stats are all that's really provided in the pstat file we get and can concretely derive formulas from. However, if you have a better way to derive ranking based on the stats that we're providing, I'd love to hear it...

[birovr]

Here I am...

I've been sharing my consideration with Ben somewhere else and he gave me good inspiration for calculating FG starting from raw stats.

First of all I would consider eFG% (effective field goal percentage' which adjusts for three-points shots made, and in the absence of three's is identical to the normal FG%) whose formula is

eFG%=((FGM + (.5 * 3PM))/FGA)

Then I would rank players according to this value using NORMAL DISTRIBUTION; I dunno hows your maths skills (but from wot Ive read above I reckon your quite good) but basically this is the "famous" bell function... It fixes a score of 50 to the average and then the more distant you are from the average the higher is your score... I learnt it form a professor at uny: he used it to avoid copying in his exams...

Or as Ben suggested me a much more easier function could be used:

player eFG%
------------------------- * 50
league eFG% average


I think next step in FG calculation should be averaging the result from eFG% with FGA (same way of calculation) - average per GAME (even if I havent decided yet if its better per game or per minute)

Here some examples (consider when I wrote this I was considering normal FG-- this should be corrected with eFG):

Bowen: FG=50 --> he scored with 0,419% (just below total players average 0,427 with stats I've loaded)... a 47 and shot a 7,3 shots per game whereas average is around 6,9... a 53.. total: 50 (with eFG% the firs part goes up)

Averaging the number of shot taken is avoiding that players that only shot once have a 99... i.e Donnell Harvey that shot a 2/2 in 3 games: he's a 99 in FG% but only a 10 in FGA (0,66 shots per game)

other examples:
Shaq: 96 ---> FG% 97 (0,601%) FGA 95 (15)
Garnett: 88 ---> FG% 80 (0,502%) FGA 97 (16,6)
D-Wade: 84 ---> FG% 71 (0,477%) FGA 98 (17,1)
Lebron: 83 ---> FG% 69 (0,472%) FGA 99 (21)

Then, I know you, HonorGlow, were looking for formula for Defensive awareness but I think it's worth to publish here "Doug's Tendex" formula (taken from www.dougstats.com, all credit to him)

"...The 'TND' fields is really just my version of the Tendex rating system. Based on the approximate Point-Per-Possession, each statistical category can be related to the game's scoring. Every time you make a basket, you have made 2 points for your team. Every basket you miss, you lose possession, and subsequently lose 1 point for your team (of course, if a team member gets an offensive rebound, that player will gain possession, and 1 TND point).

Players who have a higher TND value than their scoring average will tend to contribute to their team in ways other then scoring (i.e. high assist or rebound players). While the opposite suggets players who's job is mainly to score.

While I IN NO WAY, claim this stat to be an end-all way to rank all the players, I use it to give another data point as to how well a player is contributing to their team.

The current MyTendex formula is:

(PTS - FGmsd - FTmsd/2 + 0.5*m3 + 1.25*st + 1.25*as + bl + reb - 1.25*to - tc - 2*ff - pf/2) / GamesPlayed

..."
I dunno wot "tc" or "ff" stands for but you can easily check there...

I remember reading something about effective player rating,or something like that, a formula in stats to evaluate players according to tha stats.

To finish this one, I don't think physical ratings can be estimated... maybe strenght that could be related to weight and height of players, bu thats'it... No way to come up with realistic formula for:

DNKABILITY (think of Spud Webb)
JUMP
QUICK
SPEED
DRIBBLE

IT could be a 50*(coefficient for position) but...

Talk to you later,
b

[ParoXsitiC]

Sorry I thought you wanted to get a generated number based on ratings.
To turn real statistics into ratings is really quite difficult. Take a look at http://www.hoopsworld.com/cgi-bin/news/ ... 9&num=8293
This contains a old linear weight formula called TENDEX. It goes as follows

a*PTS + b*REB + c*AST + d*STL + e*BLK - f*TO - g*(FGA-FGM) - h*(FTA-FTM) - i*PF

where a,b,c,d... are the linear weights.

the NBA sets every weight to 1. They make every aspect of the game equally as important.

-----------------------------------
Another system you'll see in that article is VORP.
http://www.sonicscentral.com/vorp.html


=============================================

now for what you're looking for. This is an article on better methods for evaluating defense.
http://www.hoopsworld.com/cgi-bin/news/ ... 9&num=8399

Doug Steele
http://www.dougstats.com/04-05.html
John Hollinger
http://www.alleyoop.com/


This has some actual formulas:
http://www.sonicscentral.com/ratingsystem.html

--------------------------------------------------------------


Everything I just posted can all be linked to by using this page:
http://www.hoopsworld.com/cgi-bin/news/ ... 9&num=8638

[HonorGlow]

This contains a old linear weight formula called TENDEX. It goes as follows

a*PTS + b*REB + c*AST + d*STL + e*BLK - f*TO - g*(FGA-FGM) - h*(FTA-FTM) - i*PF

where a,b,c,d... are the linear weights.

the NBA sets every weight to 1. They make every aspect of the game equally as important.

Haha I'm guessing you don't visit the Dynasty threads too often…that's pretty much the Birdie formula… Yeh!!!(the formula used often by Larry Bird when he was coaching) that I've been using to evaluate many of the NLSC dynasty stories...

Also NBA.com doesn't put personal fouls as a factor…as a side correction. I have a webtool that computes both the Tendex and NBA.com formula http://www.livegmc.com/birdiefly2005.html since last October so it's good that it still has some validity.

Anyway, thanks guys so much for showing me where to find the defensive formula…I've been running into a brick wall scouring the net to find that piece of information! I will implement that into the stats analysis on my page here.