eWins inquiry thread (fundamentallysound, 2008)
Posted: Wed May 04, 2011 7:16 pm
fundamentallysound
Joined: 18 Jul 2008
Posts: 25
Location: VA
PostPosted: Tue Jul 22, 2008 9:10 am Post subject: eWins inquiry thread Reply with quote
I'm new to this message board and I've been lurking and reading for the last few days about the different metrics that people have developed and I became really jazzed about Mike G's eWins metric. So, I contacted him with my questions about it. He advised me to take these questions to a thread, so that my questions would be answered for others and in general it would be a more inclusive process. So I'm here to do just that. The questions I posed to him were these:
"What is the formula for the stat and how is it calculated? I know you calculate T first, but I'm not sure how you do that. Then you compare the player's T to R (replacement level T), but I'm also not sure how you arrive at R. Do you reset it every year or is it just a set historical value? I'm also curious how the differential between T and R (T-R) gets converted into actual eWins."
Hopefully this thread will be informative to everyone who's interested.
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Tue Jul 22, 2008 9:39 am Post subject: Reply with quote
I'm not a great communicator (to put it mildly), but I think the facts are out there. Others are good at placing links to earlier discussion; they should feel free to reproduce passages that they found enlightening.
It's not a 'proprietary' thing, but eWins kind of falls out, from a certain spreadsheet. In other words, it's hard to describe, as a single 'formula'. The data interact. A 'replacement player' is one who adds no wins. His play just treads water for a team.
'T' is a weighted sum of 7 stats, 5 of which are positive. Those 'production rates' are 'standardized' to team and opponent rates. eWins are a multiple of (T-R) x minutes. Players for good teams and bad teams have the same eW if they have the same stats; after standardizing to tm/opp PPG and RPG, that is.
These things make intuitive sense to a lot of people. Quantifying it is do-able.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
fundamentallysound
Joined: 18 Jul 2008
Posts: 25
Location: VA
PostPosted: Tue Jul 22, 2008 12:24 pm Post subject: Reply with quote
Mike thanks for the feedback I've been going through past threads that mentioned eWins (thanks to the handy search tool) and I noticed you mentioned this:
Quote:
All categories are then adjusted by MPG and % of games started; which yields a difficulty-of-opponent factor (% vs starters).
I was just curious how that was calculated. Sounds interesting.
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Tue Jul 22, 2008 5:23 pm Post subject: Reply with quote
A starter in his first part-minute is playing against 100% starters. Starters (regular season) seem to average 32 minutes. So a starter going 48 minutes will have competed vs .667 starters (32/48), on average.
A straight line (min, %St) from (0, 1.00) to (48, .667) is defined by the equation:
%St = 1 - .333*mpg/48
Starters who play few minutes have the highest % vs starters:
Code:
St% player tm GS G Min
.84 Peterson,Morris NO 76 76 24
.83 Perkins,Kendrick Bos 78 78 25
.82 Green,Willie Phi 74 74 27
.82 Curry,Eddy NY 58 59 26
.81 Fisher,Derek LAL 82 82 27
.81 Brewer,Ronnie Uta 76 76 27
.81 Krstic,Nenad NJ 38 45 18
.81 Johnson,Anthony Atl 41 42 27
.81 Haywood,Brendan Was 80 80 28
.81 Milicic,Darko Mem 64 70 24
.81 Richardson,Quent NY 65 65 28
None of these players had really outstanding per-minute numbers. But they seldom got to mop up vs bench players.
The whole formula, for players who start every game, some games, or no games, is:
St% = GS/G*(1 - .333*mpg/48.5) + (1 - GS/G)*.333*(1 + mpg/48.5)
The first term covers games started, and the second covers games off the bench. The average NBA game is about 48.5 minutes.
Guys who may have gotten decent numbers vs <50% starters:
Code:
St% player tm GS G Min
.49 Delfino,Carlos Tor 0 82 24
.49 Williams,Louis Phi 0 80 23
.49 Nachbar,Bostjan NJ 1 75 22
.48 Korver,Kyle Uta 0 50 22
.49 Smith,Joe Cle 1 27 21
.49 Najera,Eduardo Den 3 78 21
.48 Daniels,Marquis Ind 1 74 21
.49 Millsap,Paul Uta 2 82 21
.49 Anthony,Joel Mia 1 24 21
.48 Farmar,Jordan LAL 0 82 21
Assuming all NBA minutes are equal, some metrics might overrate these players. If a guy is only good for 21 minutes vs half bench players, is he good as a starter?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
fundamentallysound
Joined: 18 Jul 2008
Posts: 25
Location: VA
PostPosted: Tue Jul 22, 2008 5:33 pm Post subject: Reply with quote
Thanks again for that Mike. Now, I know how your St% for starting is calculated, how exactly do you use that factor to discount the Sco, Reb, and Ast rates that make up a big chunk of T? Is it at all similar to the way you discounted Sco using the unassisted FG estimate you developed (uA%)?
Again, thanks for being so open with all of this and for all your hard work. I really like that part of this metric that it values playing starter minutes more heavily than beating up on bench guys.
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Wed Jul 23, 2008 5:37 am Post subject: Reply with quote
Since the normal St% = .667, I just create a factor :
F = (St%/.667)^e
This year, e was = .20 . So players' rates were multiplied by a range from 1.05 (KPerkins) down to about .88 (low-mpg guys who never start). The exponent has been higher and lower before, generally .15 to .25
A St% of .667 of course has a factor of 1.
The unAssisted% applies to all points, not just FG. I figure a guy creating his own shot or a guy catch-and-shooting is being assisted/not, whether he hits the FG or is fouled. The formula is complex. As with any estimate, any is better than none.
The Celtics' uA% are topped by Rondo (.57) and bottomed by Posey (.25, and .19 in playoffs). Since league-wide, uA% tends around .40, their Sco rates are adjusted by the factor:
F = (uA%/.40)^e
When e=.090, Posey gets credit for about 94% of his points; Rondo gets 103% of his. While this may seem a small adjustment, it allows Ast to have a nice weight in the 'T' formulation.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Wed Jul 23, 2008 4:34 pm Post subject: Reply with quote
Mike, I like your metric very much. I've allways thought PGs (and maybe all positions) should have some kind of position adjustment in an Assisted factor, no matter how much we might want to reward the ballhandling skill (already attached to assists and TOs). Some players (some combo guards, point forwards, etc) are not more unassisted just because they aren't the PG, or main toucher, when an assigned PG is already in the lineup.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Thu Jul 24, 2008 9:04 am Post subject: Reply with quote
It would be a shame to create artificial 'blocks' of adjustment based on supposed position. However, of all players averaging 24+ mpg, the top 14 (and 32 of the top 40) are listed as PG (by dougstats). The clear leader was TJ Ford, who I estimated at .76, and according to 82games.com was actually even higher, at .83 (shown as 17% assisted).
I don't know that anyone's tracking 'assisted FT'; so some of the difference might be that FT are more or less 'assisted' than FG, depending on the player.
The highest uA% (estimated) among players other than PG are .60 for Ginobili (.55 actual) and Kobe (.60): Your so-called combo guards. The estimate is positively correlated to TO, Ast, Stl, FTA; negatively to FTA/FGA.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Thu Jul 24, 2008 9:50 am Post subject: Reply with quote
Well, however they (PGs) suffer "diminishing return" regarding TOs, it's fair they could compensate in an uA%factor. And an assisted factor based on individual is however an upgrade from other metrics's based on team/league average (no matter if you don't use this "team" to adjust).
Back to top
View user's profile Send private message
John Hollinger
Joined: 14 Feb 2005
Posts: 175
PostPosted: Thu Jul 24, 2008 10:00 am Post subject: Reply with quote
I tracked "assisted free throws" at games I attended in person, just for giggles, and was surprised how rare they were. I thought there might be games where a PG had seven assists but seven "foul assists", but nobody ever had more than two in a single game
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Fri Jul 25, 2008 5:02 am Post subject: Reply with quote
Well, there are 7.5 FGA for every 2-shot FT attempt, so I'd expect many more assisted FG, as well as 'potential assists' on FGA.
There is a factor based on team, which I didn't mention above. It's the player's fraction of team assists. Here's the actual formula:
uA% = ((.47*FTA + 2.06*Ast + .77*Stl + 2.9*TO)/Min - .103*3FGA/FGA) * ((TmAPG-PlaAPG)/TmAPG)*1.2
When I made this formula, I expected OffReb to create unassisted FG; but I found no correlation, apparently. It's been a while.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
mtamada
Joined: 28 Jan 2005
Posts: 376
PostPosted: Fri Jul 25, 2008 8:29 am Post subject: Reply with quote
John Hollinger wrote:
I tracked "assisted free throws" at games I attended in person, just for giggles, and was surprised how rare they were. I thought there might be games where a PG had seven assists but seven "foul assists", but nobody ever had more than two in a single game
Is it possible that you were stingier than the official scorer was/would've been? I.e. did you also compare, for actual FGMs, the assists that you would've awarded to what the official scorer awarded?
I suppose that part of it is that FTAs are relatively rare compared to FGAs. A typical superstar will attempt, what, maybe 20 FGAs per game and maybe 8 FTAs per game, which is equivalent to only 3.5 or 3.6 FGAs.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Sat Jul 26, 2008 9:04 am Post subject: Reply with quote
(pulled from a parallel thread):
Harold Almonte wrote:
I don't know how eWins measures defense
One of the units that adds into T is the standardized scoring rate (Sco). A player's points-per-36 is multiplied by 100/OppPPG; thus it's a per-100-opponent-points rate, as well.
All else being equal (St%, TS%, mpg) a player who averages 20 pts in 100-95 games is effectively scoring about 10% more than if he gets 20 PPG in 100-105 games. Whether it's a pace difference (poss/G) or a defensive effect, he's scoring a more significant part of the points in the lower-scoring environment.
On a better defensive team, the scorer is credited with more points. Then he gets credit for his steals and blocks. He gets more credit for his rebounds if his team outrebounds its opponents. My Reb rate is basically:
Reb = reb/36 * 44/OppRPG
(see also St% adjustment)
So the Reb rate is per 44 'available' (not going to one's own team) rebounds. This is a slightly higher than modern rate (42.0 in '08)
I don't distinguish between O and D rebounds. Playing style/position dictate the rebound opportunity to an extent. And having better rebounders on your team leaves you with fewer opps. Scaling to OppReb fixes much of this.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Sat Jul 26, 2008 9:53 am Post subject: Reply with quote
I'm glad to know you someway really rate rebounding, and avoid "team level likelihoods" for them (at least individually each type of rebounds), although you don't scale players's rebounding to their opportunities or reb. usages (position adjust). But, it's a win rating, not a skills rating. However, I think that like any other boxscore metric the weightest part of the defensive (1st. factor) relies on def. rebounds and its link with defFGX. Any solution for a distribution of points allowed, at least by minutes?
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Sat Jul 26, 2008 1:05 pm Post subject: Reply with quote
Regarding 'position adjustments', while it's true that interior players get more rebound opportunities -- thus more 'defensive credit' -- the perimeter guys are more likely to get first shot at a scoring attempt. That may be why eWins shows equal contributions from all positions, league-wide.
Minutes, it seems, should be included in defensive status. eWins only captures production, and scales it to the team's defense. Good defenders share their value with teammates, as it stands. But by being an ace defender, Bruce Bowen accumulates infinitely more points, rebounds, etc, than he would if he weren't such a defender.
_________________
`
36% of all statistics are wrong
gabefarkas
Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC
PostPosted: Tue Jul 29, 2008 11:07 am Post subject: Reply with quote
Mike G wrote:
Well, there are 7.5 FGA for every 2-shot FT attempt, so I'd expect many more assisted FG, as well as 'potential assists' on FGA.
There is a factor based on team, which I didn't mention above. It's the player's fraction of team assists. Here's the actual formula:
uA% = ((.47*FTA + 2.06*Ast + .77*Stl + 2.9*TO)/Min - .103*3FGA/FGA) * ((TmAPG-PlaAPG)/TmAPG)*1.2
When I made this formula, I expected OffReb to create unassisted FG; but I found no correlation, apparently. It's been a while.
I'm curious how you came to this. Was it just a straight regression with those variables? Some other method?
Back to top
View user's profile Send private message Send e-mail AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3625
Location: Hendersonville, NC
PostPosted: Tue Jul 29, 2008 4:51 pm Post subject: Reply with quote
Nothing straight about it. There are a vast number of 'interactive' variables I didn't try, in addition to '3FGA/FGA'. I just bumped the multipliers higher or lower, in order to make the estimates (league totals) closest to the uAst% kindly provided by the 82games.com guy, Roland Beech.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Joined: 18 Jul 2008
Posts: 25
Location: VA
PostPosted: Tue Jul 22, 2008 9:10 am Post subject: eWins inquiry thread Reply with quote
I'm new to this message board and I've been lurking and reading for the last few days about the different metrics that people have developed and I became really jazzed about Mike G's eWins metric. So, I contacted him with my questions about it. He advised me to take these questions to a thread, so that my questions would be answered for others and in general it would be a more inclusive process. So I'm here to do just that. The questions I posed to him were these:
"What is the formula for the stat and how is it calculated? I know you calculate T first, but I'm not sure how you do that. Then you compare the player's T to R (replacement level T), but I'm also not sure how you arrive at R. Do you reset it every year or is it just a set historical value? I'm also curious how the differential between T and R (T-R) gets converted into actual eWins."
Hopefully this thread will be informative to everyone who's interested.
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Tue Jul 22, 2008 9:39 am Post subject: Reply with quote
I'm not a great communicator (to put it mildly), but I think the facts are out there. Others are good at placing links to earlier discussion; they should feel free to reproduce passages that they found enlightening.
It's not a 'proprietary' thing, but eWins kind of falls out, from a certain spreadsheet. In other words, it's hard to describe, as a single 'formula'. The data interact. A 'replacement player' is one who adds no wins. His play just treads water for a team.
'T' is a weighted sum of 7 stats, 5 of which are positive. Those 'production rates' are 'standardized' to team and opponent rates. eWins are a multiple of (T-R) x minutes. Players for good teams and bad teams have the same eW if they have the same stats; after standardizing to tm/opp PPG and RPG, that is.
These things make intuitive sense to a lot of people. Quantifying it is do-able.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
fundamentallysound
Joined: 18 Jul 2008
Posts: 25
Location: VA
PostPosted: Tue Jul 22, 2008 12:24 pm Post subject: Reply with quote
Mike thanks for the feedback I've been going through past threads that mentioned eWins (thanks to the handy search tool) and I noticed you mentioned this:
Quote:
All categories are then adjusted by MPG and % of games started; which yields a difficulty-of-opponent factor (% vs starters).
I was just curious how that was calculated. Sounds interesting.
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Tue Jul 22, 2008 5:23 pm Post subject: Reply with quote
A starter in his first part-minute is playing against 100% starters. Starters (regular season) seem to average 32 minutes. So a starter going 48 minutes will have competed vs .667 starters (32/48), on average.
A straight line (min, %St) from (0, 1.00) to (48, .667) is defined by the equation:
%St = 1 - .333*mpg/48
Starters who play few minutes have the highest % vs starters:
Code:
St% player tm GS G Min
.84 Peterson,Morris NO 76 76 24
.83 Perkins,Kendrick Bos 78 78 25
.82 Green,Willie Phi 74 74 27
.82 Curry,Eddy NY 58 59 26
.81 Fisher,Derek LAL 82 82 27
.81 Brewer,Ronnie Uta 76 76 27
.81 Krstic,Nenad NJ 38 45 18
.81 Johnson,Anthony Atl 41 42 27
.81 Haywood,Brendan Was 80 80 28
.81 Milicic,Darko Mem 64 70 24
.81 Richardson,Quent NY 65 65 28
None of these players had really outstanding per-minute numbers. But they seldom got to mop up vs bench players.
The whole formula, for players who start every game, some games, or no games, is:
St% = GS/G*(1 - .333*mpg/48.5) + (1 - GS/G)*.333*(1 + mpg/48.5)
The first term covers games started, and the second covers games off the bench. The average NBA game is about 48.5 minutes.
Guys who may have gotten decent numbers vs <50% starters:
Code:
St% player tm GS G Min
.49 Delfino,Carlos Tor 0 82 24
.49 Williams,Louis Phi 0 80 23
.49 Nachbar,Bostjan NJ 1 75 22
.48 Korver,Kyle Uta 0 50 22
.49 Smith,Joe Cle 1 27 21
.49 Najera,Eduardo Den 3 78 21
.48 Daniels,Marquis Ind 1 74 21
.49 Millsap,Paul Uta 2 82 21
.49 Anthony,Joel Mia 1 24 21
.48 Farmar,Jordan LAL 0 82 21
Assuming all NBA minutes are equal, some metrics might overrate these players. If a guy is only good for 21 minutes vs half bench players, is he good as a starter?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
fundamentallysound
Joined: 18 Jul 2008
Posts: 25
Location: VA
PostPosted: Tue Jul 22, 2008 5:33 pm Post subject: Reply with quote
Thanks again for that Mike. Now, I know how your St% for starting is calculated, how exactly do you use that factor to discount the Sco, Reb, and Ast rates that make up a big chunk of T? Is it at all similar to the way you discounted Sco using the unassisted FG estimate you developed (uA%)?
Again, thanks for being so open with all of this and for all your hard work. I really like that part of this metric that it values playing starter minutes more heavily than beating up on bench guys.
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Wed Jul 23, 2008 5:37 am Post subject: Reply with quote
Since the normal St% = .667, I just create a factor :
F = (St%/.667)^e
This year, e was = .20 . So players' rates were multiplied by a range from 1.05 (KPerkins) down to about .88 (low-mpg guys who never start). The exponent has been higher and lower before, generally .15 to .25
A St% of .667 of course has a factor of 1.
The unAssisted% applies to all points, not just FG. I figure a guy creating his own shot or a guy catch-and-shooting is being assisted/not, whether he hits the FG or is fouled. The formula is complex. As with any estimate, any is better than none.
The Celtics' uA% are topped by Rondo (.57) and bottomed by Posey (.25, and .19 in playoffs). Since league-wide, uA% tends around .40, their Sco rates are adjusted by the factor:
F = (uA%/.40)^e
When e=.090, Posey gets credit for about 94% of his points; Rondo gets 103% of his. While this may seem a small adjustment, it allows Ast to have a nice weight in the 'T' formulation.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Wed Jul 23, 2008 4:34 pm Post subject: Reply with quote
Mike, I like your metric very much. I've allways thought PGs (and maybe all positions) should have some kind of position adjustment in an Assisted factor, no matter how much we might want to reward the ballhandling skill (already attached to assists and TOs). Some players (some combo guards, point forwards, etc) are not more unassisted just because they aren't the PG, or main toucher, when an assigned PG is already in the lineup.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Thu Jul 24, 2008 9:04 am Post subject: Reply with quote
It would be a shame to create artificial 'blocks' of adjustment based on supposed position. However, of all players averaging 24+ mpg, the top 14 (and 32 of the top 40) are listed as PG (by dougstats). The clear leader was TJ Ford, who I estimated at .76, and according to 82games.com was actually even higher, at .83 (shown as 17% assisted).
I don't know that anyone's tracking 'assisted FT'; so some of the difference might be that FT are more or less 'assisted' than FG, depending on the player.
The highest uA% (estimated) among players other than PG are .60 for Ginobili (.55 actual) and Kobe (.60): Your so-called combo guards. The estimate is positively correlated to TO, Ast, Stl, FTA; negatively to FTA/FGA.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Thu Jul 24, 2008 9:50 am Post subject: Reply with quote
Well, however they (PGs) suffer "diminishing return" regarding TOs, it's fair they could compensate in an uA%factor. And an assisted factor based on individual is however an upgrade from other metrics's based on team/league average (no matter if you don't use this "team" to adjust).
Back to top
View user's profile Send private message
John Hollinger
Joined: 14 Feb 2005
Posts: 175
PostPosted: Thu Jul 24, 2008 10:00 am Post subject: Reply with quote
I tracked "assisted free throws" at games I attended in person, just for giggles, and was surprised how rare they were. I thought there might be games where a PG had seven assists but seven "foul assists", but nobody ever had more than two in a single game
Back to top
View user's profile Send private message AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Fri Jul 25, 2008 5:02 am Post subject: Reply with quote
Well, there are 7.5 FGA for every 2-shot FT attempt, so I'd expect many more assisted FG, as well as 'potential assists' on FGA.
There is a factor based on team, which I didn't mention above. It's the player's fraction of team assists. Here's the actual formula:
uA% = ((.47*FTA + 2.06*Ast + .77*Stl + 2.9*TO)/Min - .103*3FGA/FGA) * ((TmAPG-PlaAPG)/TmAPG)*1.2
When I made this formula, I expected OffReb to create unassisted FG; but I found no correlation, apparently. It's been a while.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
mtamada
Joined: 28 Jan 2005
Posts: 376
PostPosted: Fri Jul 25, 2008 8:29 am Post subject: Reply with quote
John Hollinger wrote:
I tracked "assisted free throws" at games I attended in person, just for giggles, and was surprised how rare they were. I thought there might be games where a PG had seven assists but seven "foul assists", but nobody ever had more than two in a single game
Is it possible that you were stingier than the official scorer was/would've been? I.e. did you also compare, for actual FGMs, the assists that you would've awarded to what the official scorer awarded?
I suppose that part of it is that FTAs are relatively rare compared to FGAs. A typical superstar will attempt, what, maybe 20 FGAs per game and maybe 8 FTAs per game, which is equivalent to only 3.5 or 3.6 FGAs.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Sat Jul 26, 2008 9:04 am Post subject: Reply with quote
(pulled from a parallel thread):
Harold Almonte wrote:
I don't know how eWins measures defense
One of the units that adds into T is the standardized scoring rate (Sco). A player's points-per-36 is multiplied by 100/OppPPG; thus it's a per-100-opponent-points rate, as well.
All else being equal (St%, TS%, mpg) a player who averages 20 pts in 100-95 games is effectively scoring about 10% more than if he gets 20 PPG in 100-105 games. Whether it's a pace difference (poss/G) or a defensive effect, he's scoring a more significant part of the points in the lower-scoring environment.
On a better defensive team, the scorer is credited with more points. Then he gets credit for his steals and blocks. He gets more credit for his rebounds if his team outrebounds its opponents. My Reb rate is basically:
Reb = reb/36 * 44/OppRPG
(see also St% adjustment)
So the Reb rate is per 44 'available' (not going to one's own team) rebounds. This is a slightly higher than modern rate (42.0 in '08)
I don't distinguish between O and D rebounds. Playing style/position dictate the rebound opportunity to an extent. And having better rebounders on your team leaves you with fewer opps. Scaling to OppReb fixes much of this.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte
Joined: 04 Aug 2006
Posts: 616
PostPosted: Sat Jul 26, 2008 9:53 am Post subject: Reply with quote
I'm glad to know you someway really rate rebounding, and avoid "team level likelihoods" for them (at least individually each type of rebounds), although you don't scale players's rebounding to their opportunities or reb. usages (position adjust). But, it's a win rating, not a skills rating. However, I think that like any other boxscore metric the weightest part of the defensive (1st. factor) relies on def. rebounds and its link with defFGX. Any solution for a distribution of points allowed, at least by minutes?
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC
PostPosted: Sat Jul 26, 2008 1:05 pm Post subject: Reply with quote
Regarding 'position adjustments', while it's true that interior players get more rebound opportunities -- thus more 'defensive credit' -- the perimeter guys are more likely to get first shot at a scoring attempt. That may be why eWins shows equal contributions from all positions, league-wide.
Minutes, it seems, should be included in defensive status. eWins only captures production, and scales it to the team's defense. Good defenders share their value with teammates, as it stands. But by being an ace defender, Bruce Bowen accumulates infinitely more points, rebounds, etc, than he would if he weren't such a defender.
_________________
`
36% of all statistics are wrong
gabefarkas
Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC
PostPosted: Tue Jul 29, 2008 11:07 am Post subject: Reply with quote
Mike G wrote:
Well, there are 7.5 FGA for every 2-shot FT attempt, so I'd expect many more assisted FG, as well as 'potential assists' on FGA.
There is a factor based on team, which I didn't mention above. It's the player's fraction of team assists. Here's the actual formula:
uA% = ((.47*FTA + 2.06*Ast + .77*Stl + 2.9*TO)/Min - .103*3FGA/FGA) * ((TmAPG-PlaAPG)/TmAPG)*1.2
When I made this formula, I expected OffReb to create unassisted FG; but I found no correlation, apparently. It's been a while.
I'm curious how you came to this. Was it just a straight regression with those variables? Some other method?
Back to top
View user's profile Send private message Send e-mail AIM Address
Mike G
Joined: 14 Jan 2005
Posts: 3625
Location: Hendersonville, NC
PostPosted: Tue Jul 29, 2008 4:51 pm Post subject: Reply with quote
Nothing straight about it. There are a vast number of 'interactive' variables I didn't try, in addition to '3FGA/FGA'. I just bumped the multipliers higher or lower, in order to make the estimates (league totals) closest to the uAst% kindly provided by the 82games.com guy, Roland Beech.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail