Posted: Fri Aug 24, 2007 9:55 pm Post subject: Offensive Composite Score
Just thought I'd let everyone know that my article on Offensive Composite Score was just published at 82games. It's calculated identically to Defensive Composite Score.
4 teams without 2 guys 20+ on OCS- Cavs, Lakers (with 7th best team offensive efficiency), Wolves with just 1 and Pacers with none.
Mavs with 5 over 20 and surprisingly Bobcats did too (despite 27th ranked team offense). Mavs' 5 were in fact all over 30 to lead league on that cut with Spurs in 2nd with 4 and a handful at 3.
The top players scores are certainly interesting but depth matters in the team efficiency rankings. Sonics one of 4 teams with 3 40s but had team offensive efficiency of 12th due to injuries and poor ranks of rest of team while the other three teams were all top 4.
13 teams witn a player with an OCS 40+ made playoffs but 10 didn't. A scoring star, especially alone, is not enough.
I believe Larry Hughes was highest paid player with a negative OCS followed by Kaman, Croshere, Magliore, PJ Brown and Kurt Thomas. Defensive score helps balance the case to some degree except for Magliore and Croshere.
Nocioni's rating of 0 didnt prevent a good payday. Pietrus' negative has so far. Desmond Mason was fortunate Milwaukee offered what they did.
Bargnani not near Morrison (-51) but -20 for #1 pick is bad He got away without much heat with a low profile and a winning team but will have to improve.
I count some GM mistakes in the guys mentioned by name and some more on the teams with big offensive names who didnt make playoffs.
Last edited by Mountain on Sat Aug 25, 2007 1:09 pm; edited 1 time in total
OCS/DCS represent combined metrics. (Aaron's draft position value study also uses a combined metric approach.)
Look forward to seeing total player performance composite ratings next? I am encouraged by this, though weighting will always be a point of discussion.
Although OCS/DCS is cast as a player rating, the goodness of fit of combined metrics (of this formulation or any of an endless range of mixes) with team wins in comparison with inside/simple metrics and single outside/new metrics would be interesting to include and score in a fuller explanatory power of methods study. Maybe the outside/new metrics aren't too fancy, maybe they need to be taken to a higher level of refinement. If they have blemishes then either try to adjust the single metric (instead of leaving it as is) or thru blending and weighting it might also be possible to get closer to the desired goal. Scored this way, the end soundness after all the calculations and twists of a combined metric could be assessed and there would be a basis for re-examining and changing the weights (or components) and improving the fit. And then seeing how well the metric and the weights do with other seasons past or future and try to get a metric that is consistently high quality at explaining team results thru time. Not getting to perfect but trying to get closer.
I was thinking about what an offensive composite score would contain. In fact, I was thinking earlier today it would be cool to come up with and offensive composite score and use both OCS and DCS as methods to help compare teams, predict games, and use that as a basis to make money betting om basketball (I'm not a ref so I can do that).
I applaud the effort here. I like the inclusion of +/- and Dean Oliver's offensive score (which I don't know much about). I dislike the inclusion of PER. I don't view it as an offensive stat. Its certainly influenced by rebounds, steals, blocks, and fouls, all none of which affect a player's offensive ability/impact (except maybe offensive rebounds). PER counterpart works for the defensive version because the defensive aspects of PER even themselves out across opponents, but I don't think PER works here.
In my opinion, our own box score stat category or gategories are needed, and IMO due to each category's imperfections I think its ideal to include a total of four categories. I would propose two box score stat categories - one designed to measure a player's ability to make the player's around him more efficient. As a starting point, I would say this category would place a large primary emphasis on assists (adjusted slightly on a per team basis), turnovers, and then a lesser emphasis on three point shooting (ability to spread the D), and offensive rebounding (keeping possessions alive).
The other would measure scoring ability and instinct. Ideally I would propose that somewhere around 65% of the formula was measured by points scored multiplied by points per shot squared. Then, percentage of shots blocked (adjusted slightly per team) would count for 12.5%, FT% would count for 12.5%, and 'clutch eFG%' counted for 10%.
Then, if putting the formula together myself, as I tried to do with my own version of DCS, I'd adjust the +/- ratings per team, and adjust final scores based on minutes played overall rather than minutes played versus starters. But thats where I arguably get unnecessarily critical.
I was thinking about what an offensive composite score would contain. In fact, I was thinking earlier today it would be cool to come up with and offensive composite score and use both OCS and DCS as methods to help compare teams, predict games, and use that as a basis to make money betting om basketball (I'm not a ref so I can do that).
I applaud the effort here. I like the inclusion of +/- and Dean Oliver's offensive score (which I don't know much about). I dislike the inclusion of PER. I don't view it as an offensive stat. Its certainly influenced by rebounds, steals, blocks, and fouls, all none of which affect a player's offensive ability/impact (except maybe offensive rebounds). PER counterpart works for the defensive version because the defensive aspects of PER even themselves out across opponents, but I don't think PER works here.
In my opinion, our own box score stat category or gategories are needed, and IMO due to each category's imperfections I think its ideal to include a total of four categories. I would propose two box score stat categories - one designed to measure a player's ability to make the player's around him more efficient. As a starting point, I would say this category would place a large primary emphasis on assists (adjusted slightly on a per team basis), turnovers, and then a lesser emphasis on three point shooting (ability to spread the D), and offensive rebounding (keeping possessions alive).
The other would measure scoring ability and instinct. Ideally I would propose that somewhere around 65% of the formula was measured by points scored multiplied by points per shot squared. Then, percentage of shots blocked (adjusted slightly per team) would count for 12.5%, FT% would count for 12.5%, and 'clutch eFG%' counted for 10%.
Then, if putting the formula together myself, as I tried to do with my own version of DCS, I'd adjust the +/- ratings per team, and adjust final scores based on minutes played overall rather than minutes played versus starters. But thats where I arguably get unnecessarily critical.
I agree with your thoughts about PER. I wish I had just the offensive aspects of PER (and I may know how to calculate it). Right now my goal is getting DCS and OCS as is for as many years in the past as possible. Once I get all that data, I'll then begin revising the statistic again.
Well I went ahead, grabbed data from Dougstats' and Jon's article, and put together my own offensive composite scores for 06-07.
+/- rank accounted for 25 points. I adjusted the rank on a per team basis as I did for my defensive composite scores.
ORTG accounted for 25 points.
Assists per minute (slightly adjusted per team) accounted for 8 points. Turnovers per minute (slightly adjusted per team) 4 points, Assist/Turnover ratio 4 points, offensive rebounds per minute 4 points, 3 pt FG% 4 pts (with criteria in place to punish those with very few attempts, like Elton Brand, who went 1-1 from downtown last year). I tried to make sure 3 pt % and offensive rebounds weren't emphasized too much, as they are already arguably overemphasized in ORTG.
Points per minute multiplied by points per shot squared counted for 21 points. Free throw percentage counted for 4 points.
Then each player's score was adjusted according to minutes/games played in the same manner I adjusted my DCS scores.
Interstingly, there are three Pistons in the top 25. I view that as encouraging, since the Pistons looked weak according to my DCS scores - these OCS scores would however help explain their very good regular season record.
I then went to calculate "Overall Composite Scores." (I'll add here that since the last time I posted my DCS spreadsheet, I slightly tweaked the scores - simply by more accurately ranking DRTG, and tweaking the way PGs scores were adjusted)
For PGs, the overall score consisted of: 60% offensive score; 20% defensive score; 20% PER (offense is more important than defense @ PG, as evidenced by Steve Nash. Likewise, defense is more important @ C, as evidenced by Ben Wallace)
For SGs: 48% offensive score; 32% defensive; 20% PER rank.
For SFs: 42.2% offensive score; 37.8% defensive; 20% PER rank.
For PFs: 36% offensive score; 44% defensive; 20% PER rank.
For Cs: 24% offensive score; 56% defensive; 20% PER rank.
Then, after initially calculating scores, I felt that players averaging more than 36 minutes per game should have been rewarded slightly and proportionately. So I did that. I also adjusted some players' positions (i.e. Barbosa, who is more of a SG than PG based on his role). I am undecided as to whether Gilbert Arenas' position (and related score criteria) should be switched to SG. If Arenas is considered a SG, he drops to 83.15 and #25 in the league overall. Actually, I just decided that makes sense, so I'm editing the rest of the post accordingly.
In many player's cases I'm inclined to feel that PER should not have been included. In some player's cases however, it seems to make the results more accurate. In any event, it only accounted for about 20% of each player's score.
Interesting stuff.
On OCS would you consider a inside shot or inside FG% component to reward those players similar to what you did with 3 pt shooters? They are floor spacers and arguably help team FG% too. The current OCS top list seems heavily populated by 3 pt shooters with few post players.
Overall Composite Score is good to see arrrive. The by position ranking weights seem worth discussion. A 82games by position study I recently cited suggested that SG and SF defense was more strongly correlated with winning than the other positions and the remaining PG and bigs were about equal.
http://www.82games.com/comm31.htm
Just thinking about how defense works I'd be similarly tempted to give inside players getting a higher defensive weight but this study makes me want to go heavier on the perimeter guys than you did to reduce the size of the differential or even go back to equal weights unless the case for more heavy on the interior is buttressed.
On OCS would you consider a inside shot or inside FG% component to reward those players similar to what you did with 3 pt shooters? They are floor spacers and arguably help team FG% too. The current OCS top list seems heavily populated by 3 pt shooters with few post players.
I'd consider that but unless someone's paying me, I'm too lazy to compile that kind of data in to a spreadsheet (maybe someone has it in a spreadsheet?). Ideally I guess I'd love to be able to take player's "crunch" eFG% and possibly reward players for solid playoff performances. I'd also like to include a measure of % of shots blocked - for example last year acc to 82games, Lawrence Roberts had about 28% of his "close" i.e. layup attempts blocked, and 14% of his dunk attempts blocked. He should be punished a bit for that.
As far as the list being heavily populated by three point shooters - part of this is due to ORTG I believe - Fred Hoiberg led the league in this category two years ago. Anyways, my box score categories didn't weight three point shooting in itself heavily at all - just assists and efficiency - and points per shot.
And, I gave this some thought, guards and other perimeter players are more likely to be more athletic than bigs, more likely to be better shooters than bigs, are more likely to create offense for others than bigs because they have better driving ability, etcetera....plus, three point shots are worth more than two point shots. Just as bigs are more likely to score better on defense because they have a more suitable physique and are capable of anchoring a defense, guards are more likely to have an offensive impact. Its what I would assume at least. If you look at the overall ratings, they are not so dominated by outside shooters.
Quote:
Overall Composite Score is good to see arrrive. The by position ranking weights seem worth discussion. A 82games by position study I recently cited suggested that SG and SF defense was more strongly correlated with winning than the other positions and the remaining PG and bigs were about equal.
http://www.82games.com/comm31.htm
It is interesting and I had come across that sometime back. But I shyed away from letting it influence ratings because, well first of all, it was only based on two years of data, and the data differed in one or two cases significantly between each year. Perhaps one reason the first chart shows SF+SG defensive performance being correlated with the most wins is because there is more defensive depth @ SF/SG than any other position (that might or might not be the case). Its also worth noting that PF/Cs can easily impact production from wings by a.) providing help defense and b.) shutting down post game which frees up open shots for wings.
Cs being weighted heavily towards defense and PGs being weighted heavily towards offense seems natural to me - not just because of Ben Wallace, Steve Nash, Mark Blount when he plays good offense...but there are respected backup Cs like Joel Pryzbilla that play solid D but don't play a lick of offense. Its tough to find the opposite. Likewise there are many backup PGs that are respectable offensively but are very much incapable on the defensive end. You won't find much of the opposite.
Anyways, two other factors that were considered - PFs are likely to spend at least some time as Centers, SGs are likely to serve some time @ PG. And, all things equal, offense might be worth weighting slightly higher than D, as one could argue that its easier to teach defense than offense.
I'm thinking its possible I could have had free throw attempts count for something in the box score categories - its somewhat incorporated into points per shot, but the ability to get opposition into foul trouble should count for something aside from a player's ability to hit free throw's. On another hand though, I'm not sure its necessary, and is probably already factored a little bit into ORTG
As you say wing depth is involved. SG has the highest scoring average for league starters. SF though is 4th of the 5. Wings probably contributed a higher share of the bench scoring than th eother spots.
That 2 yr study is brief. But is that considered beforeor th ebeginning of the major change in handcheck interpretation? More recent data might make perimeter defense even more important now.
Apart from covering their own man, perimeter players role in doubling down on post play or even helping on drives may get less credit than big help defense in the lane.
Apart from covering their own man, perimeter players role in doubling down on post play or even helping on drives may get less credit than big help defense in the lane.
Yes,
though,
As the overall SG rankings stand right now for example, if anything, I'd say Ginobli looks a little overrated and Kobe & DWade look maybe slightly underrated. But if you weight defense and offense more equally, Ginobli is ranked higher while Kobe and DWade are ranked lower.
The current distribution of offensive score to defensive score makes enough logical sense to me while making the rankings look accurate enough. The more I think about it the more I'm okay with the current distributions.
There were several players assigned to the wrong position IMO. I have changed that and the new spreadsheet is here -
With T-Mac counted as a SF, he dips to #14 overall, and Steve Nash becomes a top 10 player.
San Antonio, Dallas, and Phoenix are the only teams with three players in the top 25. Spurs and Suns each have two players in the top 10, Dallas has Josh Howard @ #13. Is this a prerequisite to be a contender? Thats a reason why it would/will/might be interesting to run this historically.
Who is close? T-Mac and Yao should both be in the top 25 next year (Yao was 28 this year, hurt by missed games). Battier could improve from #47 with Adelman coacing the Rockets, but I don't see him becoming top 25. And neither Yao or TMac are a shoe in for top 10.
Chicago has two guys in the top 25 and Deng should continue to improve. But Gordon is their next best @ #53. And they have no front court post offense. Especially with Thomas registering so well defensively, perhaps they should try and trade Ben Wallace and Ben Gordon for JO and a SG. That would give them three in the top 25 (provided JO stays top 25, I usually consider him overrated, but he's in the top 25 of my calculations....though its worth noting he slips to #35 if considered a PF, and 82games' stats would indicate he's a better defender @ PF), more post offense, and they'd still have sufficient depth.
C's (I'm a C's fan)....KG at the top of the list would bode well. PP should have a much better year next year, but I'm not sure if he scores much above 80. Ray Allen is also around 80, that actually puts Ray Allen & PP in the #28-35 range (but perhaps they'll post better than expected offensive scores on a team with better offensive depth, which could put them in the top 25). So we might only have one player in the top 25, even if we would bode well to win the East in lieu of a good trade by the Bulls (statistics aside the Bulls don't worry me without a good trade executed anyway).
While I don't expect Toronto to build much on their success next season for certain reasons, they're not too far away from having 3 in the top 25. But Chris Bosh is their highest ranked player @ 18, and has current issues with planar fascitis. Also Anthony Parker @28 and Calderon @44 (should start perhaps?) aren't guys that have huge "upsides."
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
