Introducing eWins (MikeG, 2005)
Posted: Wed Apr 20, 2011 3:58 am
Author Message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Fri Mar 04, 2005 11:28 am Post subject: introducing eWins Reply with quote
Expected Wins, Equivalent Wins, I don't know what to call this thing.
I took my 2004 file, which rates players, and subtracted a replacement-level value (equivalent to an average 12th-man), and apportioned eWins to every player. These do not add up to team Wins, unless it's a 41-win team. But so far, it looks like you can just about double the separation from 41 and get actual (schedule-adjusted, pythagorean-expected) team wins.
So a team whose players add to 30 wins is not really going to win 30; being 11 games under .500, just double that shortfall and expect 19 wins (41 - 22) from that lineup.
How well does this predict the success of a team that's changed it's roster from last year? Well, there are major complicating factors: Rookies, injuries, major player improvements, etc.
Let's start at the top with the defending champ Pistons. An intact starting lineup, and wholesale changes to the bench. Currently they're turning their season around; still, they've had no major injuries or midseason trades. Here's their lineup with eWins from last year, and from this year (scaled to 82 games):
2004 eWins.. 2005 eWins
9.6 RWallace 7.3
9.5 Billups ... 9.1
8.4 Hamilton 8.6
7.4 BWallace 7.3
5.9 Arroyo ... 2.9
5.2 Prince .... 6.7
1.5 Mcdyess . 5.3
1.5 Coleman 0.0
0.8 Dupree .. 0.3
0.5 Hunter ... 1.2
---------------
50.3 .. Total .. 48.2
The player minutes from last year add up almost exactly to (240x82) so don't need adjustment. The Predictor column shows 50 wins, and they're on course for 53. Part of that, at least, would be due to the fact they're in the East. It shouldn't be too hard to find average wins to add or subtract depending on the conference.
The 2nd column might suggest they're better than their record indicates.
If I double the distance from 41 wins, they should win 55 games.
Sorry in advance for any mistakes or omissions. This monster is just a baby.
Back to top
View user's profile Send private message Send e-mail
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Tue Mar 08, 2005 11:48 am Post subject: Reply with quote
Well, this thread is uncrowded, to say the least. A few breakthroughs on my end, since last post:
I've (tentatively) determined the value of a replacement player. That's someone who doesn't actually contribute any wins to an average team. The only reason to play such a guy is to see what he's got, in garbage minutes. Projects and over-the-hill players.
The value of the replacement player is 11.75, in my 'system'. There have been 377 players better than this level, this year: about 13 per team. Not everyone is available at any given time (injuries, basically), so I guess that's intuitively acceptable.
Another offshoot of my revamping is a complete overhaul of Weights I'll be assigning to various conventional stat categories. These are now correlated to eWins -- which as part of their definition is "over replacement value".
For the curious, here are my before-and-after weights. "Scoring" is directly scaled by (TS%/.527), where .527 is historic average TS%. Sco, Reb, and Ast are scaled to Team and Opponent averages.
stat -- old -- new
Sco - 1.00 - 1.00
Reb - 1.00 - .895
Ast - 1.33 - 1.52
PF. - 0.25 - 0.25
Stl - 1.50 - 1.76
TO - 1.25 - 1.69
Blk - 1.50 - 2.25
All these values give the highest correlations to eWins.
Scoring is defined as 1.00, and everything else is relative to that.
Rebounds are indeed <1 in value. I cannot discern any difference in importance between OR and DR.
Assists are contentious; I always valued them >1 to justify the minutes received by high-Ast players. Plus, I like passing.
Somehow my gut guess for fouls was right-on. (PF and TO are negative.)
I upgraded steals last year due to DanR's studies. And some more just now.
Turnovers get more weight, too. They're tantalizingly close to Steals.
I'd always given blocks the 1.5 value. Even that was way short.
The free throw factor I use is .45, rather than .44. But the best correlation to wins turns out to be more like .47. This suggests perhaps that some FTA are "forced", and there's negative repercussions from "no-calls".
The standard TS% of comparison that correlates best to wins is around .540, (not the historic .527 nor the current .515). Maybe that's what TS% "should be", if every team got the ball to the best shooter every time.
I didn't attempt to correlate to actual season wins. Neither did I shoot for Pythagorean pt.-diff Expected Wins. Rather, I generated "equivalent wins" in a balanced league -- as if every team had the same strength of schedule -- from Sagarin's team ratings. I call these "Sagarin Wins", for now.
My mean error is less than 2, between sWins and eWins.
Back to top
View user's profile Send private message Send e-mail
kjb
Joined: 03 Jan 2005
Posts: 855
Location: Washington, DC
PostPosted: Tue Mar 08, 2005 12:31 pm Post subject: Reply with quote
Mike G wrote:
Another offshoot of my revamping is a complete overhaul of Weights I'll be assigning to various conventional stat categories. These are now correlated to eWins -- which as part of their definition is "over replacement value".
For the curious, here are my before-and-after weights. "Scoring" is directly scaled by (TS%/.527), where .527 is historic average TS%. Sco, Reb, and Ast are scaled to Team and Opponent averages.
stat -- old -- new
Sco - 1.00 - 1.00
Reb - 1.00 - .895
Ast - 1.33 - 1.52
PF. - 0.25 - 0.25
Stl - 1.50 - 1.76
TO - 1.25 - 1.69
Blk - 1.50 - 2.25
All these values give the highest correlations to eWins.
Scoring is defined as 1.00, and everything else is relative to that.
Rebounds are indeed <1 in value. I cannot discern any difference in importance between OR and DR.
Assists are contentious; I always valued them >1 to justify the minutes received by high-Ast players. Plus, I like passing.
Somehow my gut guess for fouls was right-on. (PF and TO are negative.)
I upgraded steals last year due to DanR's studies. And some more just now.
Turnovers get more weight, too. They're tantalizingly close to Steals.
I'd always given blocks the 1.5 value. Even that was way short.
Fascinating stuff here. In the work I've been doing on defense, I've found a similar kind of correlation between blocks and defense and steals and defense.
Quote:
The free throw factor I use is .45, rather than .44. But the best correlation to wins turns out to be more like .47. This suggests perhaps that some FTA are "forced", and there's negative repercussions from "no-calls".
What do you mean here? Free throws are "forced" by the offensive player penetrating to the hoop? And how does this .45 -- .44 -- .47 difference show the effect of no-calls?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Tue Mar 08, 2005 12:48 pm Post subject: Reply with quote
Kevin, I'm just guessing. If .44 represents a ratio of scoring-attempts/FTA, then a lower ratio suggest FT are an even better deal, while a bigger number suggests some tradeoff in efficiency.
I rather expected to see a factor lower than .44, to account for the positive effect of creating foul trouble in the opposition. It seems intuitively that teams that get to the line can often make up for rebounding deficits, shooting woes, etc. Is it fully accounted for in the points (FT) scored? And then some?
All I have come up with is that over-reliance on drawing fouls might deleteriously affect individual/team performance. A war of attrition that the aggressor gets the short end of.
Alternate explanations are warmly received.
Back to top
View user's profile Send private message Send e-mail
FrontRange
Joined: 27 Jan 2005
Posts: 131
PostPosted: Tue Mar 08, 2005 12:53 pm Post subject: Reply with quote
Do blocks correlate closely to other stats?
I wouldn't be surprised that steals and blocks have such a large impact on wins, provided that what you are really capturing is above average athletic ability and basketball sense/ability.
For example Duncan, O'Neals, Camby, AK, Wallace are all very good basketball players who also get alot of blocks. However, Ostertag who tends to come up high on alot ratings isn't a very good basketball player who still gets alot of blocks. I get concerned about drawing conclusions about the values of players like that (Prizbilla, Foyle, Hunter, etc) bcs they happen to perform well in a stat that correlates well with other important basketball activities.
Back to top
View user's profile Send private message
FrontRange
Joined: 27 Jan 2005
Posts: 131
PostPosted: Tue Mar 08, 2005 12:56 pm Post subject: FT Reply with quote
YOu might try running the stats excluding Utah results . . .they are so far from the norm in FT/fouling that it might make a marginal difference.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Wed Mar 09, 2005 9:15 am Post subject: Reply with quote
I'd be leary of drawing correlations between various stats. I treat each as a free-standing attribute, for better or worse.
Ostertag was pretty good last year, but I don't mean pretty. I have him contributing 5.0 wins, which (because he was with a near-.500 team) is right between b-r.com's PW and 1/3 of WS. This year he's terrible in every way.
Utah this year puts opponents at the line more than anyone else. I've also got them "most overrated" this year. Hmm. Other high-fouling teams seem to be randomly placed on that scale.
Players that have been moved during the year I am only looking at askance. For example, Jim Jackson played big minutes in Houston and now appears with his whole year's stats in Phoenix. I'm scaling his numbers to a much higher-scoring/rebounding milieu, so it looks as if he sucks.
Overall, this year's correlations (team player-win totals to actual wins) aren't nearly as close as last year's (on which the parameters are based). Hopefully after the season (and before the postseason) I will have them sorted out, with players' part-seasons calculated separately for their various teams.
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Fri Mar 11, 2005 8:46 pm Post subject: a few comments Reply with quote
i'd be interested in seeing the "e-win" calculations for all the teams when you do the calculations.
i am still cautious about all stat formulas that are expressed in terms of wins. this is a much sought and desirable format but illusive.
every play has impact but some end up being more decisive in terms of winning than others due to timing. and there are plenty examples of teams seeming losing most of the box score but still winning the game. talent, heart and intensity dont really get incorporated into summary evaluations but some way should be found.
with only a few points separately winning from losing in many games perhaps there should be complete stats on performance in the last x (5?)minutes of the game since a large number of games are still up for grabs at that time marker.
i'd give a heavier winning weight to the stats of players and teams that perform well in the last 5 minutes.
and with past winning records (like last 10 or 20 games) i assume a pretty good indicator of future winning maybe you assign a weight based on that to. that is shaping the formula to the expected and desired results but it also may get you closer.
good luck with your continuing work on this. i'll look forward to reading more.
this stat could also be used to compare rotations and try to discern the optimal rotation for a particular game or generalized for a full season. it would be interesting to see how many solutions for e-wins can be found that beat the true winning percentage (to be fair, it should on a going forward process, current rotation vs. recommended measure in future games, not hindsight rotation "fixing" of past results. and then have a realistic discussion whether these are valid strategies or false hopes or hard to tell which glimmers that leave armchair coaches and GMS without any clearer measure of their levelof game smarts than their own humble or prideful self-assessment.
p.s. i liking passing too , but i find "assists" over given by scorers and i believe over weighted here. i know there is a lot to this argument, with assists a proxy for overall passing contributions, and dont really intend to reopen it, just offering a brief dissent on the weight looking somewhat high to me.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sat Mar 12, 2005 9:30 am Post subject: Reply with quote
Quote:
i'd be interested in seeing the "e-win" calculations for all the teams when you do the calculations.
Well, I have this year's to date. And I'm working on past seasons (to 1978). For this year, I have players grouped by their team of last appearance; so soon after the big trade flurry, this will throw off teams' real totals. Walker has "carried" his eWins to Boston, for example
Scaled to 82 games:
54.2 SA
47.9 Mia
46.9 Phe
46.3 Cle
46.0 Dal
45.4 Det
43.2 Sac
43.1 Sea
43.0 LAC
42.6 Mem
42.3 Hou
42.0 Bos
41.4 Min
40.6 Den
38.8 Ind
38.3 Phl
37.7 LAL
37.4 NJ
37.3 Por
36.8 Orl
36.5 Chi
36.4 Was
36.3 Tor
35.7 Uta
35.3 NY
33.8 Cha
32.1 NO
31.7 Mil
27.6 GS
23.4 Atl
Phoenix (46.9) has twice as much "talent" as Atlanta (23.4), above replacement level. This is crudely said, but serves as an operational definition. To deduce expected Team wins, I haven't settled firmly on a formula; but one that seems to fit very well so far, is simply to double the wins above-or-below 41.
Atlanta has players capable of contributing to wins; but because of competition, they can't win 23 games. They're 17.6 games short of average, and are only able to win (41-17.6*2 =) 6 games, with their current lineup.
I think I'll have to use a "half the distance" rule, and say half the distance between 23.4 and zero is about 12 games.
Back to top
View user's profile Send private message Send e-mail
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sat Mar 12, 2005 10:44 am Post subject: Reply with quote
Quote:
i am still cautious about all stat formulas that are expressed in terms of wins. this is a much sought and desirable format but illusive.
every play has impact but some end up being more decisive in terms of winning than others due to timing. and there are plenty examples of teams seeming losing most of the box score but still winning the game. talent, heart and intensity dont really get incorporated into summary evaluations but some way should be found.
I fully agree, and it was earlier discussions about some disparities between observation and earlier WS and PW numbers that were disquieting. I conjectured a better way, and then I pursued it.
Are Stockton and Malone really among the most "win-producing" players of all time? Or were they just great players who were on winning teams every year?
I'm entering team wins, team Pythagorean Expected wins (b-r.com), and a strength-of-schedule index, to use as correlators to eWins. One thing I'm discovering thru this methodical work is how some franchises consistently over- or under-achieve.
Between 1978 and 1986, the Pacers averaged 3 fewer wins per year than they should have -- if PythExt wins defines "should have won". Between 1980 and 1990, the Lakers won 3.5 games More per year than they should have.
In 1980, the Pistons were terrible; point-differential suggest they should have won 22 games. But they won only 16.
Do certain teams have more self-respect.? Do Magic's Lakers just exude confidence and win most of their close games? Did Dickie V's Pistons just expect to lose every night? Did the Pacers fail to get over the hump for so long just because they got no respect?
The Lakers post-Magic have continued to overachieve in the Win column. Not as consistently, but regularly. In some of their Shaq-era postseason series, they were outscored yet advanced.
If there's any merit to this pursuit of equitable "win-credits", what do we think is the real correlation to seek? Wins as recorded? Pythagorean Expected Wins? Schedule-adjusted (Sagarinlike) team strength?
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sat Mar 12, 2005 12:58 pm Post subject: more on ewins and team "competitive personality" Reply with quote
i read your reply below and the results for 2003-04 look fairly tightly fitted and the 2004-05 figures look much better now, for most teams, compared to current winning percentages projected wins for 82 games too after making the twice the distance from 41 wins adjustment. (you are right i misunderstood the ewin data presented above as being final win projections rather as the intermediate data still needing to be adjusted.)
one new thing i was wondering about is whether it would be possible and/or wise to try to adjust the data to reflect differences between teams regarding the frequency of playing in games decided by 3-5 points or less and by 10 or more points.
two teams with full season average data that look similar but have very different frequency and records in close and blowout games are probably more different than the full season average data suggests. teams with similar close and blowout records may be closer together than their full season average data suggests. you could call this their competitive personality or winning attitude or smarts. something we both recognized the need for inclusion by some means in previous posts.
i have not thought that thru fully but it might be worth pondering. my first observation is that most of the good teams have a low number of losses that are 3 points or less. i am not sure what if anything to do with number of wins and losses 10 points or more. since winning is really what's important and margin isnt, i was wondering if you might damp down the ewin score of a team whose stats are inflated by a lot of blowouts and maybe adjust upward the ewin score of a team whose average stats are perhaps more heavily hurt by higher than average number of blowouts.
i couldnt find the 2003-2004 data on close games and blowouts, so i looked at it for this year using this data
http://sports.espn.go.com/nba/expanded?season=2005
if i am calculating it correctly, cleveland has an ewin projection of 51.6 but a current winning pace of only 44.4 for a difference of 7.2 wins overprojected. the fact that they are the only team with at least their winning percentage who has a below .500 record in games decided by 3 points or less (though it is only 3-4) may mean something. that they are among the top 5 teams in number of blowouts might too. milwaukee has an ewin projection of 22.4 but a current winning percentage projection of 31.7, for a difference of 11.8 underprojected. i found that they have the second best record of teams with their win percentage or less in games decided by 3 points or less (8-6, second only to portland who are an unexpected 7-3). they also among the 5 teams with the least number of blowouts played. hanging around fairly close and having a pretty good record in close games and not playing too many blowouts seems to suggest they are better at winning than their average stats suggest.
looking at other outliers in terms of difference between ewin projection and current winning percentage, chicago (underprojected by ewins by 12.8 ) is also quite low in blowouts and ok in close games. washington (underprojected by 14) is fabulous in close games (9-5) and has moderately low number of blowouts played. charlotte (overprojected by 10) is poor in close games (4-10), and though that is still better than their overall record, it is still one of the leagues worst. atlanta has a bad close game record and very high blowout frequency. but seattle's underprojection doesnt really get clarified with this approach - they are just ok in close games (4-4) and play a lot of blowouts. however the strong performance of phoenix in close games (6-1) seems to partially explain their over performance (underprojected by ewin by 9.4).
overall this data suggest that competitive profiles might be helpful to improve ewin predictive closeness but it isnt obvious exactly how to account for it.
Last edited by jambalaya on Sat Mar 12, 2005 8:39 pm; edited 18 times in total
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sat Mar 12, 2005 4:39 pm Post subject: Reply with quote
Jambalaya,
Sorry, I was unclear. The list I posted was individual sums of eWins; to "project" team eWins, double the distance from 41.
Once that is done, West teams should appear to be slightly underrated, and East to be slightly overrated. That's because my correlations are designed to be impervious to strength of schedule. I'm trying to normalize all players regardless of team situation.
Your suggestion that statistical weights could be adjusted is just what I did. If I change them, the team-total eWins are less correlated with actual team strength. I thought you were asking about this year's totals, so I posted them, with disclaimers about player movement.
Here are the 2003-04 numbers. I'm dropping the decimal for clarity. The 3 columns are Actual Wins, "Sagarin Wins", and eWins.
tm Win sWin eWin
San 57 60 56
Min 58 56 52
Sac 55 54 51
Dal 52 53 52
Ind 61 53 52
Det 54 52 59
LAL 56 50 47
Mem 50 47 48
Hou 45 44 45
NJ 47 44 46
Den 43 44 43
Mil 41 40 42
GS 37 40 41
Sea 37 40 39
Mia 42 40 41
Uta 42 39 44
Por 41 39 43
NO 41 38 39
NY 39 36 37
Bos 36 35 34
Cle 35 34 37
Phi 33 33 35
Tor 33 33 33
Pho 29 33 32
LAC 28 32 33
Atl 28 30 30
Was 25 27 27
Chi 23 27 27
Orl 21 26 23
Teams are listed in the order they were ranked by Sagarin at the end of the year. His numbers (ranging from about 85 to 97, avg 90.0) I converted into win totals, if all schedules were equal.
Somehow my number is closer to actual wins, in many cases. Teams I had "most overrated" were Det (6.0 G), Utah (4.5) and Por (4.0). Relative to Sagarin, I most undervalued SA (-4.6), Min (-3.8), and Orl (-3.5). The average error is 1.78. The average eWins was 41.0
I actually suppose the average error is no bigger than the average difference between (Wins) and (Expected Wins by the Pythagorean formula). No, I'm not entirely satisfied; I really think we should get closer to perfection here. Looking at past teams, there seems to be an evolution of some of the parameters. I'll be tweaking them.
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sat Mar 12, 2005 8:33 pm Post subject: my calculation of ewin projections Reply with quote
using the twice distance from 41 formula suggested this is what i get for this year's data and what i used in the discussion in my previous post
Atlantic c win% c proj. wins ind. sum ewin proj diff
Boston 0.532 43.6 42 43.0 -0.6
Phila 0.484 39.7 38.3 35.6 -4.1
New Jer 0.435 35.7 37.4 33.8 -1.9
New York 0.433 35.5 35.3 29.6 -5.9
Toronto 0.419 34.4 36.3 31.6 -2.8
Central Pct
Detroit 0.617 50.6 45.4 49.8 -0.8
Cleveland 0.542 44.4 46.3 51.6 7.2
Chicago 0.542 44.4 36.5 32.0 -12.4
Indiana 0.508 41.7 38.8 36.6 -5.1
Milwaukee 0.417 34.2 31.7 22.4 -11.8
Southeast Pct
Miami 0.746 61.2 47.9 54.8 -6.4
Wash 0.559 45.8 36.4 31.8 -14.0
Orlando 0.508 41.7 36.8 32.6 -9.1
Charlotte 0.203 16.6 33.8 26.6 10.0
Atlanta 0.18 14.8 23.4 5.8 -9.0
Western Conference Standings
Southwest Pct
San Ant 0.77 63.1 54.2 67.4 4.3
Dallas 0.656 53.8 46 51.0 -2.8
Houston 0.59 48.4 42.3 43.6 -4.8
Memphis 0.574 47.1 42.6 44.2 -2.9
New Orl 0.217 17.8 32.1 23.2 5.4
Northwest Pct
Seattle 0.683 56.0 43.1 45.2 -10.8
Denver 0.517 42.4 40.6 40.2 -2.2
Minnesota 0.508 41.7 41.4 41.8 0.1
Portland 0.373 30.6 37.3 33.6 3.0
Utah 0.328 26.9 35.7 30.4 3.5
Pacific Pct
Phoenix 0.758 62.2 46.9 52.8 -9.4
Sacra 0.619 50.8 43.2 45.4 -5.4
LA Lakers 0.517 42.4 37.7 34.4 -8.0
LA Clipper0.435 35.7 43 45.0 9.3
Golden St 0.306 25.1 27.6 14.2 -10.9
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sun Mar 13, 2005 7:28 am Post subject: Reply with quote
Wow! Thanks for taking the time to grasp this concept, and for the corroborating work. As you can see, in the East, only the Cavs and Cats are being overestimated. Then you go out and find they're terrible in the close games.
Charlotte's 4-10 record in close games almost completely explains their "overrating" in my system. That's a small sample; and just a handful of plays going their way gives them 4-5 more wins: 6-8 more projected to a full season.
Sometime in the last couple of years, Toronto was winning all their nail-biters, and putting up a nice W-L mark. But they were still a lousy team, posing as a decent one. Their luck ran out before long.
The Cavs are led by a superstar who is just a kid. I don't think I'll go there and tag Lebron as an "uncompetetive personality". Rather, I'll say the Cavs are better than their record.
Still I get your point, and when in such situations I'll likely do what I usually do: split the difference. That is, take some point between Wins and Pyth-Wins, and see how everything else correlates.
Glancing across your revised chart of "projected" eWins, I think my numbers are a few days old. They don't really add up to 41 x 30 wins, do they? Even in the West, many teams are underestimated. I guess I only posted the list to rank teams in order.
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sun Mar 13, 2005 2:49 pm Post subject: lebron Reply with quote
i suspect he will quickly learn how to get his team to play even better in the close games. if they dont lose z, i would expect them to live up much closer to their ewin projection next year.
p.s. you're welcome on the feedback given and thanks for the return
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Fri Mar 04, 2005 11:28 am Post subject: introducing eWins Reply with quote
Expected Wins, Equivalent Wins, I don't know what to call this thing.
I took my 2004 file, which rates players, and subtracted a replacement-level value (equivalent to an average 12th-man), and apportioned eWins to every player. These do not add up to team Wins, unless it's a 41-win team. But so far, it looks like you can just about double the separation from 41 and get actual (schedule-adjusted, pythagorean-expected) team wins.
So a team whose players add to 30 wins is not really going to win 30; being 11 games under .500, just double that shortfall and expect 19 wins (41 - 22) from that lineup.
How well does this predict the success of a team that's changed it's roster from last year? Well, there are major complicating factors: Rookies, injuries, major player improvements, etc.
Let's start at the top with the defending champ Pistons. An intact starting lineup, and wholesale changes to the bench. Currently they're turning their season around; still, they've had no major injuries or midseason trades. Here's their lineup with eWins from last year, and from this year (scaled to 82 games):
2004 eWins.. 2005 eWins
9.6 RWallace 7.3
9.5 Billups ... 9.1
8.4 Hamilton 8.6
7.4 BWallace 7.3
5.9 Arroyo ... 2.9
5.2 Prince .... 6.7
1.5 Mcdyess . 5.3
1.5 Coleman 0.0
0.8 Dupree .. 0.3
0.5 Hunter ... 1.2
---------------
50.3 .. Total .. 48.2
The player minutes from last year add up almost exactly to (240x82) so don't need adjustment. The Predictor column shows 50 wins, and they're on course for 53. Part of that, at least, would be due to the fact they're in the East. It shouldn't be too hard to find average wins to add or subtract depending on the conference.
The 2nd column might suggest they're better than their record indicates.
If I double the distance from 41 wins, they should win 55 games.
Sorry in advance for any mistakes or omissions. This monster is just a baby.
Back to top
View user's profile Send private message Send e-mail
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Tue Mar 08, 2005 11:48 am Post subject: Reply with quote
Well, this thread is uncrowded, to say the least. A few breakthroughs on my end, since last post:
I've (tentatively) determined the value of a replacement player. That's someone who doesn't actually contribute any wins to an average team. The only reason to play such a guy is to see what he's got, in garbage minutes. Projects and over-the-hill players.
The value of the replacement player is 11.75, in my 'system'. There have been 377 players better than this level, this year: about 13 per team. Not everyone is available at any given time (injuries, basically), so I guess that's intuitively acceptable.
Another offshoot of my revamping is a complete overhaul of Weights I'll be assigning to various conventional stat categories. These are now correlated to eWins -- which as part of their definition is "over replacement value".
For the curious, here are my before-and-after weights. "Scoring" is directly scaled by (TS%/.527), where .527 is historic average TS%. Sco, Reb, and Ast are scaled to Team and Opponent averages.
stat -- old -- new
Sco - 1.00 - 1.00
Reb - 1.00 - .895
Ast - 1.33 - 1.52
PF. - 0.25 - 0.25
Stl - 1.50 - 1.76
TO - 1.25 - 1.69
Blk - 1.50 - 2.25
All these values give the highest correlations to eWins.
Scoring is defined as 1.00, and everything else is relative to that.
Rebounds are indeed <1 in value. I cannot discern any difference in importance between OR and DR.
Assists are contentious; I always valued them >1 to justify the minutes received by high-Ast players. Plus, I like passing.
Somehow my gut guess for fouls was right-on. (PF and TO are negative.)
I upgraded steals last year due to DanR's studies. And some more just now.
Turnovers get more weight, too. They're tantalizingly close to Steals.
I'd always given blocks the 1.5 value. Even that was way short.
The free throw factor I use is .45, rather than .44. But the best correlation to wins turns out to be more like .47. This suggests perhaps that some FTA are "forced", and there's negative repercussions from "no-calls".
The standard TS% of comparison that correlates best to wins is around .540, (not the historic .527 nor the current .515). Maybe that's what TS% "should be", if every team got the ball to the best shooter every time.
I didn't attempt to correlate to actual season wins. Neither did I shoot for Pythagorean pt.-diff Expected Wins. Rather, I generated "equivalent wins" in a balanced league -- as if every team had the same strength of schedule -- from Sagarin's team ratings. I call these "Sagarin Wins", for now.
My mean error is less than 2, between sWins and eWins.
Back to top
View user's profile Send private message Send e-mail
kjb
Joined: 03 Jan 2005
Posts: 855
Location: Washington, DC
PostPosted: Tue Mar 08, 2005 12:31 pm Post subject: Reply with quote
Mike G wrote:
Another offshoot of my revamping is a complete overhaul of Weights I'll be assigning to various conventional stat categories. These are now correlated to eWins -- which as part of their definition is "over replacement value".
For the curious, here are my before-and-after weights. "Scoring" is directly scaled by (TS%/.527), where .527 is historic average TS%. Sco, Reb, and Ast are scaled to Team and Opponent averages.
stat -- old -- new
Sco - 1.00 - 1.00
Reb - 1.00 - .895
Ast - 1.33 - 1.52
PF. - 0.25 - 0.25
Stl - 1.50 - 1.76
TO - 1.25 - 1.69
Blk - 1.50 - 2.25
All these values give the highest correlations to eWins.
Scoring is defined as 1.00, and everything else is relative to that.
Rebounds are indeed <1 in value. I cannot discern any difference in importance between OR and DR.
Assists are contentious; I always valued them >1 to justify the minutes received by high-Ast players. Plus, I like passing.
Somehow my gut guess for fouls was right-on. (PF and TO are negative.)
I upgraded steals last year due to DanR's studies. And some more just now.
Turnovers get more weight, too. They're tantalizingly close to Steals.
I'd always given blocks the 1.5 value. Even that was way short.
Fascinating stuff here. In the work I've been doing on defense, I've found a similar kind of correlation between blocks and defense and steals and defense.
Quote:
The free throw factor I use is .45, rather than .44. But the best correlation to wins turns out to be more like .47. This suggests perhaps that some FTA are "forced", and there's negative repercussions from "no-calls".
What do you mean here? Free throws are "forced" by the offensive player penetrating to the hoop? And how does this .45 -- .44 -- .47 difference show the effect of no-calls?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Tue Mar 08, 2005 12:48 pm Post subject: Reply with quote
Kevin, I'm just guessing. If .44 represents a ratio of scoring-attempts/FTA, then a lower ratio suggest FT are an even better deal, while a bigger number suggests some tradeoff in efficiency.
I rather expected to see a factor lower than .44, to account for the positive effect of creating foul trouble in the opposition. It seems intuitively that teams that get to the line can often make up for rebounding deficits, shooting woes, etc. Is it fully accounted for in the points (FT) scored? And then some?
All I have come up with is that over-reliance on drawing fouls might deleteriously affect individual/team performance. A war of attrition that the aggressor gets the short end of.
Alternate explanations are warmly received.
Back to top
View user's profile Send private message Send e-mail
FrontRange
Joined: 27 Jan 2005
Posts: 131
PostPosted: Tue Mar 08, 2005 12:53 pm Post subject: Reply with quote
Do blocks correlate closely to other stats?
I wouldn't be surprised that steals and blocks have such a large impact on wins, provided that what you are really capturing is above average athletic ability and basketball sense/ability.
For example Duncan, O'Neals, Camby, AK, Wallace are all very good basketball players who also get alot of blocks. However, Ostertag who tends to come up high on alot ratings isn't a very good basketball player who still gets alot of blocks. I get concerned about drawing conclusions about the values of players like that (Prizbilla, Foyle, Hunter, etc) bcs they happen to perform well in a stat that correlates well with other important basketball activities.
Back to top
View user's profile Send private message
FrontRange
Joined: 27 Jan 2005
Posts: 131
PostPosted: Tue Mar 08, 2005 12:56 pm Post subject: FT Reply with quote
YOu might try running the stats excluding Utah results . . .they are so far from the norm in FT/fouling that it might make a marginal difference.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Wed Mar 09, 2005 9:15 am Post subject: Reply with quote
I'd be leary of drawing correlations between various stats. I treat each as a free-standing attribute, for better or worse.
Ostertag was pretty good last year, but I don't mean pretty. I have him contributing 5.0 wins, which (because he was with a near-.500 team) is right between b-r.com's PW and 1/3 of WS. This year he's terrible in every way.
Utah this year puts opponents at the line more than anyone else. I've also got them "most overrated" this year. Hmm. Other high-fouling teams seem to be randomly placed on that scale.
Players that have been moved during the year I am only looking at askance. For example, Jim Jackson played big minutes in Houston and now appears with his whole year's stats in Phoenix. I'm scaling his numbers to a much higher-scoring/rebounding milieu, so it looks as if he sucks.
Overall, this year's correlations (team player-win totals to actual wins) aren't nearly as close as last year's (on which the parameters are based). Hopefully after the season (and before the postseason) I will have them sorted out, with players' part-seasons calculated separately for their various teams.
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Fri Mar 11, 2005 8:46 pm Post subject: a few comments Reply with quote
i'd be interested in seeing the "e-win" calculations for all the teams when you do the calculations.
i am still cautious about all stat formulas that are expressed in terms of wins. this is a much sought and desirable format but illusive.
every play has impact but some end up being more decisive in terms of winning than others due to timing. and there are plenty examples of teams seeming losing most of the box score but still winning the game. talent, heart and intensity dont really get incorporated into summary evaluations but some way should be found.
with only a few points separately winning from losing in many games perhaps there should be complete stats on performance in the last x (5?)minutes of the game since a large number of games are still up for grabs at that time marker.
i'd give a heavier winning weight to the stats of players and teams that perform well in the last 5 minutes.
and with past winning records (like last 10 or 20 games) i assume a pretty good indicator of future winning maybe you assign a weight based on that to. that is shaping the formula to the expected and desired results but it also may get you closer.
good luck with your continuing work on this. i'll look forward to reading more.
this stat could also be used to compare rotations and try to discern the optimal rotation for a particular game or generalized for a full season. it would be interesting to see how many solutions for e-wins can be found that beat the true winning percentage (to be fair, it should on a going forward process, current rotation vs. recommended measure in future games, not hindsight rotation "fixing" of past results. and then have a realistic discussion whether these are valid strategies or false hopes or hard to tell which glimmers that leave armchair coaches and GMS without any clearer measure of their levelof game smarts than their own humble or prideful self-assessment.
p.s. i liking passing too , but i find "assists" over given by scorers and i believe over weighted here. i know there is a lot to this argument, with assists a proxy for overall passing contributions, and dont really intend to reopen it, just offering a brief dissent on the weight looking somewhat high to me.
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sat Mar 12, 2005 9:30 am Post subject: Reply with quote
Quote:
i'd be interested in seeing the "e-win" calculations for all the teams when you do the calculations.
Well, I have this year's to date. And I'm working on past seasons (to 1978). For this year, I have players grouped by their team of last appearance; so soon after the big trade flurry, this will throw off teams' real totals. Walker has "carried" his eWins to Boston, for example
Scaled to 82 games:
54.2 SA
47.9 Mia
46.9 Phe
46.3 Cle
46.0 Dal
45.4 Det
43.2 Sac
43.1 Sea
43.0 LAC
42.6 Mem
42.3 Hou
42.0 Bos
41.4 Min
40.6 Den
38.8 Ind
38.3 Phl
37.7 LAL
37.4 NJ
37.3 Por
36.8 Orl
36.5 Chi
36.4 Was
36.3 Tor
35.7 Uta
35.3 NY
33.8 Cha
32.1 NO
31.7 Mil
27.6 GS
23.4 Atl
Phoenix (46.9) has twice as much "talent" as Atlanta (23.4), above replacement level. This is crudely said, but serves as an operational definition. To deduce expected Team wins, I haven't settled firmly on a formula; but one that seems to fit very well so far, is simply to double the wins above-or-below 41.
Atlanta has players capable of contributing to wins; but because of competition, they can't win 23 games. They're 17.6 games short of average, and are only able to win (41-17.6*2 =) 6 games, with their current lineup.
I think I'll have to use a "half the distance" rule, and say half the distance between 23.4 and zero is about 12 games.
Back to top
View user's profile Send private message Send e-mail
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sat Mar 12, 2005 10:44 am Post subject: Reply with quote
Quote:
i am still cautious about all stat formulas that are expressed in terms of wins. this is a much sought and desirable format but illusive.
every play has impact but some end up being more decisive in terms of winning than others due to timing. and there are plenty examples of teams seeming losing most of the box score but still winning the game. talent, heart and intensity dont really get incorporated into summary evaluations but some way should be found.
I fully agree, and it was earlier discussions about some disparities between observation and earlier WS and PW numbers that were disquieting. I conjectured a better way, and then I pursued it.
Are Stockton and Malone really among the most "win-producing" players of all time? Or were they just great players who were on winning teams every year?
I'm entering team wins, team Pythagorean Expected wins (b-r.com), and a strength-of-schedule index, to use as correlators to eWins. One thing I'm discovering thru this methodical work is how some franchises consistently over- or under-achieve.
Between 1978 and 1986, the Pacers averaged 3 fewer wins per year than they should have -- if PythExt wins defines "should have won". Between 1980 and 1990, the Lakers won 3.5 games More per year than they should have.
In 1980, the Pistons were terrible; point-differential suggest they should have won 22 games. But they won only 16.
Do certain teams have more self-respect.? Do Magic's Lakers just exude confidence and win most of their close games? Did Dickie V's Pistons just expect to lose every night? Did the Pacers fail to get over the hump for so long just because they got no respect?
The Lakers post-Magic have continued to overachieve in the Win column. Not as consistently, but regularly. In some of their Shaq-era postseason series, they were outscored yet advanced.
If there's any merit to this pursuit of equitable "win-credits", what do we think is the real correlation to seek? Wins as recorded? Pythagorean Expected Wins? Schedule-adjusted (Sagarinlike) team strength?
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sat Mar 12, 2005 12:58 pm Post subject: more on ewins and team "competitive personality" Reply with quote
i read your reply below and the results for 2003-04 look fairly tightly fitted and the 2004-05 figures look much better now, for most teams, compared to current winning percentages projected wins for 82 games too after making the twice the distance from 41 wins adjustment. (you are right i misunderstood the ewin data presented above as being final win projections rather as the intermediate data still needing to be adjusted.)
one new thing i was wondering about is whether it would be possible and/or wise to try to adjust the data to reflect differences between teams regarding the frequency of playing in games decided by 3-5 points or less and by 10 or more points.
two teams with full season average data that look similar but have very different frequency and records in close and blowout games are probably more different than the full season average data suggests. teams with similar close and blowout records may be closer together than their full season average data suggests. you could call this their competitive personality or winning attitude or smarts. something we both recognized the need for inclusion by some means in previous posts.
i have not thought that thru fully but it might be worth pondering. my first observation is that most of the good teams have a low number of losses that are 3 points or less. i am not sure what if anything to do with number of wins and losses 10 points or more. since winning is really what's important and margin isnt, i was wondering if you might damp down the ewin score of a team whose stats are inflated by a lot of blowouts and maybe adjust upward the ewin score of a team whose average stats are perhaps more heavily hurt by higher than average number of blowouts.
i couldnt find the 2003-2004 data on close games and blowouts, so i looked at it for this year using this data
http://sports.espn.go.com/nba/expanded?season=2005
if i am calculating it correctly, cleveland has an ewin projection of 51.6 but a current winning pace of only 44.4 for a difference of 7.2 wins overprojected. the fact that they are the only team with at least their winning percentage who has a below .500 record in games decided by 3 points or less (though it is only 3-4) may mean something. that they are among the top 5 teams in number of blowouts might too. milwaukee has an ewin projection of 22.4 but a current winning percentage projection of 31.7, for a difference of 11.8 underprojected. i found that they have the second best record of teams with their win percentage or less in games decided by 3 points or less (8-6, second only to portland who are an unexpected 7-3). they also among the 5 teams with the least number of blowouts played. hanging around fairly close and having a pretty good record in close games and not playing too many blowouts seems to suggest they are better at winning than their average stats suggest.
looking at other outliers in terms of difference between ewin projection and current winning percentage, chicago (underprojected by ewins by 12.8 ) is also quite low in blowouts and ok in close games. washington (underprojected by 14) is fabulous in close games (9-5) and has moderately low number of blowouts played. charlotte (overprojected by 10) is poor in close games (4-10), and though that is still better than their overall record, it is still one of the leagues worst. atlanta has a bad close game record and very high blowout frequency. but seattle's underprojection doesnt really get clarified with this approach - they are just ok in close games (4-4) and play a lot of blowouts. however the strong performance of phoenix in close games (6-1) seems to partially explain their over performance (underprojected by ewin by 9.4).
overall this data suggest that competitive profiles might be helpful to improve ewin predictive closeness but it isnt obvious exactly how to account for it.
Last edited by jambalaya on Sat Mar 12, 2005 8:39 pm; edited 18 times in total
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sat Mar 12, 2005 4:39 pm Post subject: Reply with quote
Jambalaya,
Sorry, I was unclear. The list I posted was individual sums of eWins; to "project" team eWins, double the distance from 41.
Once that is done, West teams should appear to be slightly underrated, and East to be slightly overrated. That's because my correlations are designed to be impervious to strength of schedule. I'm trying to normalize all players regardless of team situation.
Your suggestion that statistical weights could be adjusted is just what I did. If I change them, the team-total eWins are less correlated with actual team strength. I thought you were asking about this year's totals, so I posted them, with disclaimers about player movement.
Here are the 2003-04 numbers. I'm dropping the decimal for clarity. The 3 columns are Actual Wins, "Sagarin Wins", and eWins.
tm Win sWin eWin
San 57 60 56
Min 58 56 52
Sac 55 54 51
Dal 52 53 52
Ind 61 53 52
Det 54 52 59
LAL 56 50 47
Mem 50 47 48
Hou 45 44 45
NJ 47 44 46
Den 43 44 43
Mil 41 40 42
GS 37 40 41
Sea 37 40 39
Mia 42 40 41
Uta 42 39 44
Por 41 39 43
NO 41 38 39
NY 39 36 37
Bos 36 35 34
Cle 35 34 37
Phi 33 33 35
Tor 33 33 33
Pho 29 33 32
LAC 28 32 33
Atl 28 30 30
Was 25 27 27
Chi 23 27 27
Orl 21 26 23
Teams are listed in the order they were ranked by Sagarin at the end of the year. His numbers (ranging from about 85 to 97, avg 90.0) I converted into win totals, if all schedules were equal.
Somehow my number is closer to actual wins, in many cases. Teams I had "most overrated" were Det (6.0 G), Utah (4.5) and Por (4.0). Relative to Sagarin, I most undervalued SA (-4.6), Min (-3.8), and Orl (-3.5). The average error is 1.78. The average eWins was 41.0
I actually suppose the average error is no bigger than the average difference between (Wins) and (Expected Wins by the Pythagorean formula). No, I'm not entirely satisfied; I really think we should get closer to perfection here. Looking at past teams, there seems to be an evolution of some of the parameters. I'll be tweaking them.
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sat Mar 12, 2005 8:33 pm Post subject: my calculation of ewin projections Reply with quote
using the twice distance from 41 formula suggested this is what i get for this year's data and what i used in the discussion in my previous post
Atlantic c win% c proj. wins ind. sum ewin proj diff
Boston 0.532 43.6 42 43.0 -0.6
Phila 0.484 39.7 38.3 35.6 -4.1
New Jer 0.435 35.7 37.4 33.8 -1.9
New York 0.433 35.5 35.3 29.6 -5.9
Toronto 0.419 34.4 36.3 31.6 -2.8
Central Pct
Detroit 0.617 50.6 45.4 49.8 -0.8
Cleveland 0.542 44.4 46.3 51.6 7.2
Chicago 0.542 44.4 36.5 32.0 -12.4
Indiana 0.508 41.7 38.8 36.6 -5.1
Milwaukee 0.417 34.2 31.7 22.4 -11.8
Southeast Pct
Miami 0.746 61.2 47.9 54.8 -6.4
Wash 0.559 45.8 36.4 31.8 -14.0
Orlando 0.508 41.7 36.8 32.6 -9.1
Charlotte 0.203 16.6 33.8 26.6 10.0
Atlanta 0.18 14.8 23.4 5.8 -9.0
Western Conference Standings
Southwest Pct
San Ant 0.77 63.1 54.2 67.4 4.3
Dallas 0.656 53.8 46 51.0 -2.8
Houston 0.59 48.4 42.3 43.6 -4.8
Memphis 0.574 47.1 42.6 44.2 -2.9
New Orl 0.217 17.8 32.1 23.2 5.4
Northwest Pct
Seattle 0.683 56.0 43.1 45.2 -10.8
Denver 0.517 42.4 40.6 40.2 -2.2
Minnesota 0.508 41.7 41.4 41.8 0.1
Portland 0.373 30.6 37.3 33.6 3.0
Utah 0.328 26.9 35.7 30.4 3.5
Pacific Pct
Phoenix 0.758 62.2 46.9 52.8 -9.4
Sacra 0.619 50.8 43.2 45.4 -5.4
LA Lakers 0.517 42.4 37.7 34.4 -8.0
LA Clipper0.435 35.7 43 45.0 9.3
Golden St 0.306 25.1 27.6 14.2 -10.9
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3429
Location: Delphi, Indiana
PostPosted: Sun Mar 13, 2005 7:28 am Post subject: Reply with quote
Wow! Thanks for taking the time to grasp this concept, and for the corroborating work. As you can see, in the East, only the Cavs and Cats are being overestimated. Then you go out and find they're terrible in the close games.
Charlotte's 4-10 record in close games almost completely explains their "overrating" in my system. That's a small sample; and just a handful of plays going their way gives them 4-5 more wins: 6-8 more projected to a full season.
Sometime in the last couple of years, Toronto was winning all their nail-biters, and putting up a nice W-L mark. But they were still a lousy team, posing as a decent one. Their luck ran out before long.
The Cavs are led by a superstar who is just a kid. I don't think I'll go there and tag Lebron as an "uncompetetive personality". Rather, I'll say the Cavs are better than their record.
Still I get your point, and when in such situations I'll likely do what I usually do: split the difference. That is, take some point between Wins and Pyth-Wins, and see how everything else correlates.
Glancing across your revised chart of "projected" eWins, I think my numbers are a few days old. They don't really add up to 41 x 30 wins, do they? Even in the West, many teams are underestimated. I guess I only posted the list to rank teams in order.
Back to top
View user's profile Send private message Send e-mail
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sun Mar 13, 2005 2:49 pm Post subject: lebron Reply with quote
i suspect he will quickly learn how to get his team to play even better in the close games. if they dont lose z, i would expect them to live up much closer to their ewin projection next year.
p.s. you're welcome on the feedback given and thanks for the return