Having drafted their center of the future, Robert Swift, the Sonics looked to move center Calvin Booth for a rebounder who could help address the team's weakness on the boards. With only a few clicks of their mouse, Walker (now president and CEO), GM Rick Sund or Assistant GM Rich Cho could call up a list of all the NBA players they could trade Booth for straight up who were above their minimum rate of rebounds per 48 minutes.
From this list, one name jumped out: Dallas forward/center Danny Fortson, whose contract was virtually identical to Booth's and who had led the league in rebounds per 48 minutes (19.2) in 2003-04. The Sonics made a deal for Fortson, and midway through his first season in Seattle, he's averaging 8.9 points and 6.2 rebounds per game and has played a key role in the Sonics emergence as one of the NBA's top rebounding teams this season.
Were these scenarios what Walker imagined when he told a prospective intern, Cho, in 1995 that he wanted to make the Sonics into the NBA's most technically advanced team? Yes, as were many more that have come to fruition thanks to an NBA player evaluation system unlike any that had ever been used before.
Joined: 03 Jan 2005 Posts: 497 Location: Greensboro, North Carolina
Posted: Thu Feb 03, 2005 4:24 am Post subject:
Quote:
What makes the SEN unique is that the weighting for various statistics depends on the position. When Sonics Explorer was first being built, the Sonics coaching staff and front-office personnel submitted their list of the most important statistical criteria for each position. A composite of these rankings was used to create a formula for each position. Notably, the Sonics place a heavy importance on assist-to-turnover ratios for point guards, and their ratings at the position reflect that.
There are a lot of interesting points made in this article, but this one caught my eye. It seems to suggest that assist-to-turnover ratios are more important for point guards than for other positions. That might be true but in my work, I have generally found that an extra assist or one less turnover tends to add less to the adjusted plus/minus of point guards than it does for other positions. Interestingly, it appears that an extra rebound, extra block, or extra free throw attempt adds more to the adjusted plus/minus for point guards vs. other positions.
Moreover, the point guard position is by far the hardest position to model with traditional statistics. Moreso than other positions, it appears that traditional statistics do a worse job explaining the value of point guards. My theory is this is largely because of "system assists" that are a by-product of the particular sets a team runs. A point guard playing for Charlotte this season probably is going to have a lot of assists, regardless of his true value. Even Steve Smith has generated a lot of assists when he has played point guard. On the other hand, the point guard for a team like the Kings or Spurs is probably never going to generate a lot of assists, because the offensive sets often are run through the PF position. Now it is tough to separate the system from the point guard, because teams tailor their sets to the skills of their point guards, but still I think this makes it harder to evaluate point guards (versus other positions) with traditional statistics.
I also think this is tough for non-statistical evaluation. It means that the value of point guard is very wrapped up in the offensive system he runs. Move him to a new team with a different system and his value may change a lot. Thus, a Steve Nash may be far more valuable for a team like Phoenix than he was for Dallas. A Brevin Knight is one of the league's better point guards playing for Charlotte, but in another system he may be nothing more than a mediocre point guard.
Last edited by Dan Rosenbaum on Thu Feb 03, 2005 2:33 pm; edited 1 time in total
Joined: 03 Jan 2005 Posts: 497 Location: Greensboro, North Carolina
Posted: Thu Feb 03, 2005 12:55 pm Post subject:
gabefarkas wrote:
could it be that other teams are doing this or similar things, but aren't talking about it as much?
For the vast majority of teams, no. In one way or another, DeanO, Roland, and I probably have been in touch with almost every team in the NBA about their stats operations. I would say that Dallas and San Antonio have pretty sophisticated stats operations. Philly, Miami, Houston, Denver, and Portland have reached out to the stats community from time to time. I think that Detroit, Phoenix, and Memphis each has their own in-house stats guys, but I don't know a lot about exactly what they do. But in talking wth these teams, I think it safe to rule out something like what Seattle does. And that other teams, such as Chicago, Indiana, New Orleans, Golden State, and Charlotte have some level of interest in increasing their stats capabilities.
The Spurs are a very analytical group and Sam Presti is very good with this statistical stuff. The Mavs, of course, use WINVAL, but their stats use runs much deeper than that. I recently had lunch with Mark Cuban, and he is very good stats guy himself. At one point in the conversation, he started saying that "even though they don't know it, coaches are running regressions in their head when they are evaluating players or game strategies." (This is not an exact quote.) I was pretty impressed because not a lot of people think about the world that way. He also is a damn good economist too.
Dan, I think I know the answer to this, but is DanVAL sophisticated enough to factor out that point guards with low assist/turnover ratios often tend to be better scorers?
I also wonder if the analysis might change if we limited to backup point guards, whose role is usually more about not screwing up than doing positive things. The guys the Sonics have targetted (Ollie, Daniels) have been backups.
Joined: 03 Jan 2005 Posts: 497 Location: Greensboro, North Carolina
Posted: Fri Feb 04, 2005 1:35 am Post subject:
admin wrote:
Dan, I think I know the answer to this, but is DanVAL sophisticated enough to factor out that point guards with low assist/turnover ratios often tend to be better scorers?
I also wonder if the analysis might change if we limited to backup point guards, whose role is usually more about not screwing up than doing positive things. The guys the Sonics have targetted (Ollie, Daniels) have been backups.
On the first point, I am accounting for practically everything that I can account for with traditional statistics, including scoring and shooting percentage.
On the second point, back-up point guards like Bobby Jackson, Earl Boykins, Carlos Arroyo (in Detroit), Ben Gordon and Juan Dixon (though they arguably are SGs) are PGs who provide a spark for their teams. Practically speaking, I can't throw a lot of PGs out of the analysis without risking that the results becoming too noisy to be useful.
Joined: 30 Dec 2004 Posts: 534 Location: Near Philadelphia, PA
Posted: Fri Feb 04, 2005 2:39 am Post subject:
Dan Rosenbaum wrote:
admin wrote:
Dan, I think I know the answer to this, but is DanVAL sophisticated enough to factor out that point guards with low assist/turnover ratios often tend to be better scorers?
On the first point, I am accounting for practically everything that I can account for with traditional statistics, including scoring and shooting percentage.
This raises a question actually. Do you include a code for position at all in your evaluations? _________________ Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com
Joined: 03 Jan 2005 Posts: 497 Location: Greensboro, North Carolina
Posted: Fri Feb 04, 2005 3:21 am Post subject:
HoopStudies wrote:
Dan Rosenbaum wrote:
admin wrote:
Dan, I think I know the answer to this, but is DanVAL sophisticated enough to factor out that point guards with low assist/turnover ratios often tend to be better scorers?
On the first point, I am accounting for practically everything that I can account for with traditional statistics, including scoring and shooting percentage.
This raises a question actually. Do you include a code for position at all in your evaluations?
I did not in my evaluations last Spring and Summer, but more recently I realized that the strong results I was getting for versatility were picking up differences in the parameters for players who play different roles. So now I run some of the regressions separately by position. I also run them separately by true shooting attempts.
Posted: Mon Feb 07, 2005 7:37 pm Post subject: per minute projections
the article mentions rick sund likes to get the per minute rates projected at 36 minutes. some advocate using 30 minutes.
there is sometimes problems/mistakes with projecting minutes for emerging players well beyond their actual playing time track record.
here are two possible general usage rules for per minute projections for emerging players for consideration:
1. per minute projections for a player in a current or immediately following year should generally not exceed some figure, perhaps 50% more than their current playing time average.
so for a player playing less than 10 minutes a game, you wouldnt generally project beyond 15 minutes a game. for a player getting 15 minutes, not project beyond 22.5 minutes. for a player at 20 minutes a game, not beyond 30 minutes.
2. or alternatively, you could simplify and standardize it by saying if current minutes are below 10 per game, project at some preferred figure between 15- 20 minutes; if current minutes are between 10-20, project at something like 25 minutes; if current over 20, go ahead and project to 30. i hesitate to project a bench player beyond 30 minutes.
some form of guideline like these examples would keep the projection somewhat closer to the demonstrated performance on the floor and limit the distance of the projection "leap".
of course you could make special exceptions if you feel very confident and / or have a large role to fill. at your own peril.
if you do project a low minutes guy out to 30 minutes or beyond, further than what would be called for in a % growth or step based approach, it probably should be labelled speculative or long-run. or understood clearly by all as merely a standardized score and not called a projection. some stretches seem too far to be responsibly called a projection.
These are two separate arguments. When Sund refers to 36 minutes, he means that's the number to which stats are projected, as opposed to the standard 48 minutes. (John Hollinger, who continues to remain curiously absent, uses 40 minutes.)
John's research has suggested that players tend to perform better, not worse, when given more minutes.
Posted: Tue Feb 08, 2005 12:30 am Post subject: Should Be Two
admin wrote:
These are two separate arguments.
They should be two seperate numbers / arguments. A standardized production score at some per minute level, and a projection (based on actually how much you think you might use a guy and reasonably expect to get out of him in that role). This is an underlying element of my post, it that wasn't clear at least it was a fundamental part of my thinking on the topic.
I am not convinced that this distinction is or has always been keep in mind in front offices. this article doesnt present any information about a separate player specific usage/performance expectation projection number with the Sonics. is their one? is that coming in the final installment of the series?
i would hope there are two numbers. that is why I suggested that a standardize score at one uniform level at 30, 36, 40, or 48 or whatever minutes isnt sufficent by itself for the statistical portion of the process for evaluating players and assembling a team/rotation and achieving desirable results across statistical categories.
i suggested that the projections should be different and player appropriate- i.e. more restrained in the case of bench players expanding their minutes to reflect a growth process into larger roles and more minutes, usually not going from modest to full-time all at once.
from the outside i am not conviced a separate and realistic projection of expected performance was prepared and evaluated distinct from a per minute standardized score in the cases of mcilvane and booth to name a few Sonic examples. both appear to have been signed and paid very well based on straight line extrapolations of previous year limited minute per minute production levels to bigger minutes. having watched them, i have a hard time understanding how the scouts could have confidently believed them ever capable of playing 25-30 minutes in the NBA at a solid performance level. (they just didnt seem to have the energy, strength and stamina to go to a much bigger role.) i seem to remember that their per 48 minutes stats, especially for blocks was mentioned as partial justification for the signings in conversations with the press when they were introduced. maybe they were never envisioned to be that big in minutes, but they were sure paid and talked up in a way that made it seem like they were going to be. maybe there was a reasonable and wholistic case assembled for signing them and they just disappointed. i didnt know the answer because i dont have all the inside facts. but rather than focus on them, my point is that it smelled like the early Sonic trials with using per minute stats as a significant contributing decisionmaking factor were not yet adequately calibrated or successfully integrated with scouting and cost effectiveness evaluations and the rest of it.
if you can add more detail about the actual past and current practices, that would be welcome, though i understand there are probably and appropriately constraints. i didnt necessarily expect the sonics would want to cover that old ground or admit mistakes but it could have added depth and perspective if they were. the new statistical age is an evolution, of trial and error and adjustment. i'd expect they've gotten better at it.
p.s. i havent seen the hollinger study you mention about performance being up with more minutes. while that general finding seems reasonable, i would think the data would be fairly mixed because not all players react the same way to more time and responsibility. but if you can note the link or the message post number, i would read it.
Joined: 14 Jan 2005 Posts: 1521 Location: Delphi, Indiana
Posted: Tue Feb 08, 2005 8:58 am Post subject:
Back when Chris Gatling and Will Perdue somehow "looked like" superstars from their per-minute projections, I knew something had to be done. I tried multiplying not by a factor of (36/MPG), but by the square-root of 36/MPG, and then by the 4th-root.
The square-root adjustment might be a better fit for a Perdue, who couldn't effectively go beyond 20 minutes or so. But for most players, the 4th-root might be closer to right.
Here's an example of per-36-minute "projections" of a guy who scores a point per minute, by the above-mentioned methods:
These hypothetical players would of course be projected to 36 points in 36 minutes, done linearly. As you can see, the differences are profound for low-minute players and negligible at 25-30 min.
JohnH and I (and others in A-analysis) debated the effect of increased minutes upon production. My contention is that every case is unique, and some players fade after X minutes, while others cannot be productive off the bench. So I tend to trust coaching staffs and their hands-on knowledge. That is, I assume there are valid reasons when a guy is getting limited PT.
We agreed that the best available test might be of players who suddenly get more minutes due to injury hitting a major player, and how the players filling the void produce.
But even in this circumstance, players who show improvement are going to be given more minutes. Of the several players available to take up minutes, those who do it better will get the minutes. So the cause-and-effect question arises; and I still don't think the mere act of giving minutes is likely to create better per-minute production.
Some years ago, someone did a study showing Mark Eaton's production to drop significantly after about 30 minutes in most games. Going another 6-8 minutes might still get more out of the center position, depending on his backup; but he was really just a 30-32 minute player.
When players get weary, they aren't good. A shotblocker gets more fouls and fewer blocks, because he's always catching up to the play. A shooter may slacken on defense.
I guess in all I'd say there's value in a player who can go 30-40 minutes at full bore, and there's value if that guy who can still be effective in limited minutes. Since coming off the bench usually amounts to playing against 2nd-stringers, there's some statistical reward in that job, too.
My contention is that every case is unique, and some players fade after X minutes, while others cannot be productive off the bench. So I tend to trust coaching staffs and their hands-on knowledge. That is, I assume there are valid reasons when a guy is getting limited PT.
I agree that every case is unique, but I think there are simply too many factors involved in determining playing time to assume that a player plays a certain number of minutes because that's all he can play. I don't think you can look at a player's minutes per game and determine from that how playing more or less would affect him.
Joined: 03 Jan 2005 Posts: 497 Location: Greensboro, North Carolina
Posted: Tue Feb 08, 2005 10:51 am Post subject:
I guess I don't see per-minute stats as being a problem. The per-minute stats are just a way to hold minutes constant and compare players on other dimensions. I think everyone realizes that in some cases per-minute stats will not be great indicators of what a player might do with 36 minutes a game. Some arbitrary correction (that we would argue endlessly over) is not a substitute for judgment.
That said, in my adjusted plus/minus work I have found that after accounting for traditional statisitics, minutes per game still helps explain why some players have higher adjusted plus/minus ratings than others. This is evidence that coaches, as a group, do know what they are doing. But the effect is pretty small, so it really is not a justification for many of the choices that coaches make.
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
