APBRmetrics

Author Message
Ed Küpfer



Joined: 30 Dec 2004
Posts: 764
Location: Toronto

PostPosted: Sun Apr 09, 2006 4:38 pm Post subject: Clustering Players Reply with quote
One of my hobby horses is classifying players by their roles, in contrast to their positions. I have long suspected that traditional position designations were not very useful, a relic of an earlier game much different than the one we watch today. I believe that we can do better, that we can come up with player classifications that are more useful in the context of the modern game, the same way we dumped FG% in favour of measures that better reflect what we see on the court today.

To that end, I've been playing around with cluster analysis. CA is a family of algorithms that cluster observations together automatically based solely on their stats, without regard to how we would cluster them -- that is, there is no dependant variable in CA. I won't describe here how cluster analysis works, but I will say that the logic and the math behind it is very simple, and anyone who wants to read up on it will probably grok the concepts very easily.

There are different types of cluster analysis, and one can break them into two kinds: hierarchical (HCA) and non-hierarchical (NHCA). The former returns results in the form of a tree diagram, which is really helpful. If you went to a large family reunion and ran a HCA, it would probably begin by clustering siblings together, and then clustering those siblings with their parents, and then clustering the parents/sibs cluster with other parents/sibs clusters based on the proximity of familial relations. You would end up with something like a family tree diagram. You can also run a HCA on a group of players, using their stats as input variables, and seeing how the "family tree" of players looks. I've done this many times, and essentially the tree looks like this:

Code:

|
+--------------------+-------------------+
| |
Frontcourt Backcourt
| |
+---------+----------+ +---------+----------+
| | | |
Primary offensive non-offesnive Primary offensive non-offesnive
options players options players

Kirilenko Griffin Ford RBowen
Slater Kaman Cassell Augmon
Yao Voskuhl Hudson BBowen




That is a very, very rough tree, based on only three seasons of data. It's still very useful, and maybe I'll talk more about the HCA tree later.

But what I really wanted to look at is if players' classifications changed throughout their careers. To do that, I needed to use more data. The problem is that running a hierarchical cluster analysis is very computationally intensive — it would take forever to run the HCA on the dataset I wanted to used, which included every player season from 1978 on. Luckily, statisticians have developed other algorithms which don't require as much computation. I file these under the heading non-hierarchical cluster analysis. The most well known is called k-means, where k is the number of clusters you want the computer to return. Unlike HCA, k-means doesn't settle on an optimal number of clusters — it wants you to tell it how many clusters there are. I don't really want to use this, because we don't really have an idea of how many "natural" clusters there are.

Fortunately, there are other options. SPSS has a NHCA called two-step cluster analysis. I don't really know what the two steps are, but the algorithm settles automatically on the number of "natural" clusters, which makes it very useful for me.

Right. The data. Here are the stats I used:

HT: Player height
WT: Player weight
2Att: 2-point attempts per min
3Att: 3-point attempts per min
FTA: FT attempts per min
PF: Personal fouls per min
USAGE: Usage rate
OReb : Offensive rebounding percentage
DReb : Defensive rebounding percentage
TO: Turnover percentage
AST: Percentage of teammate attempts assisted
BLK: Percentage of opponent shots blocked
STL: Steals per opponent possessions
qAST: Percentage of own attempts assisted

All stats pace adjusted to team/league averages. For players who played for more than one team in a season I used the average, weighted by the minutes played.

The cluster analysis settled on 7 clusters. I've named these clusters Post Players, Driving Swingmen, Human Victory Cigars, Miscellaneous Role Players, Defensive Specialists, Backcourt Ballhandlers, and Outside Shooters. These names are just convenient titles, capturing what I see as the clusters' most salient characteristics along with the things that separate them most from the other clusters.

Each cluster has a stats "profile." For example, the POST PLAYER cluster is characterised by high totals in defensive rebounding, two-point attempt rate, and FT attempt rate, average totals in PFs, turnovers, assists, and few 3-point attempts. Like this:

POST PLAYERS

Code:
High: DReb, 2Att, FTA, WT, HT, USAGE, OReb, qAST, BLK
Avg: PF, TO, STL, AST
Low: 3Att
Eddy Curry, Dirk Nowitzki, Drew Gooden, Stromile Swift, Juwan Howard, Zendon Hamilton, Rasheed Wallace, Patrick Ewing, Elton Brand, Lamar Odom

The ten players' at the end are drawn randomly from the POST cluster. This cluster is probably the most intuitively satisfying one. Here are the remaining clusters.

DRIVING SWINGMEN

Code:
High: 2Att, USAGE, FTA
Avg: STL, qAST, AST, 3Att, OReb, HT
Low: DReb, WT, PF, BLK, TO
Lebron James, Jeryl Sasser, Allen Iverson, Ronald Murray, Ricky Davis, Richard Jefferson, James Cotton, Isiah Rider, Kobe Bryant, Ron Mercer

You won't see Jeryl Sasser's name appear next to Iverson's too often, but I think we can visualise this type of player easily enough.


HUMAN VICTORY CIGARS

Code:
High: PF, TO, STL, FTA, USAGE
Avg: OReb, qAST, 2Att, 3Att, HT, WT, DReb, BLK
Low: AST
Lawrence Funderburke, Tim James, Rusty LaRue, Terry Mills, Jermaine Jackson, Rashard Lewis, Tierre Brown, Damone Brown, Jason Hart, Jerome James

This is the most diffuse cluster, containing players who you'd think have very little in common. In fact, at the end of this post, I'll show a map of all these clusters, and while the others have pretty well defined borders and territories, the CIGARS are in fact all over the place. The one thing they share unambiguously is a high games played/minutes per game ratio, a diagnostic I've used before to flag garbage time players. The players in this cluster played an average of 5 minutes per game, far lower than the second lowest (DEFENSIVE SPECIALISTS - 15 mpg). It's important to remember that MPG was not a stat I used as an input. The clustering algorithm classified these garbage time players without knowing their playing time ahead of time. To me, this is a good external verification of the existence of this cluster.

In know what you're thinking: what the hell is Rashard Lewis doing on a list of garbage time players? All I can say is that Lewis is represented here by his 1999 season, when he only played 145 minutes. His full career trajectory looks like this: CIGAR, ROLE PLAYER, ROLE PLAYER, SHOOTER, DRIVER, SHOOTER, SHOOTER.


MISC ROLE PLAYERS

Code:
High: qAST
Avg: HT, OReb, WT, PF, DReb, STL, TO, BLK, 3Att, 2Att, FTA
Low: USAGE, AST
Michael Curry, Marcus Haislip, Kenny Thomas, Detlef Schrempf, Jonathan Bender, Kevin Edwards, Robert Horry, Carlos Rogers, Ansu Sesay, Vincent Yarbrough

The "miscellaneous" is apt, I think. These players are defined by their inability to create much in the way of offense, but are otherwise average in other stats. Other than the cigars, this is the cluster that contains the greatest range of player positions.

Proportion of ROLE PLAYERS from each traditional player position:

Code:
PG 1%
G 5%
SG 6%
GF 16%
SF 14%
F 28%
PF 16%
FC 13%
C 2%

The characteristic that separates the players in this cluster from the CIGARS is that these players get much more action. These players obviously have some ability, although it doesn't show up much in the stats I used.

DEFENSIVE SPECIALISTS

Code:
High: WT, HT, BLK, DReb, OReb, PF, qAST
Avg: TO, FTA
Low: USAGE, 3Att, AST, STL, 2Att
Charles Oakley, Jahidi White, Clarence Weatherspoon, Dennis Rodman, Hakeem Olajuwon, Jackson Vroman, Joe Kleine, Rasho Nesterovic, Maciej Lampe, Reggie Slater

I could also have called this cluster DEFENSIVE BIG MEN. The similarities between these players are pretty obvious: PFs and centers, lots of rebounds, lots of fouls, few assists and field goal attempts. Very straightforward. That said, Reggie Slater? I loved him on Saved By The Bell, but from his days with the Raptors, I don't remember him playing much defense.


BACKCOURT BALLHANDLERS

Code:
High: AST, STL, TO
Avg: 3Att, USAGE, FTA, 2Att
Low: qAST, HT, WT, OReb, DReb, BLK, PF
Chris Childs, Kevin Ollie, Allen Iverson, Keyon Dooling, Charlie Ward, Will Avery, Speedy Claxton, Tony Parker, Mike James, Kenny Anderson
As you'll see in the map below, the BALLHANDLERS are closely related to the DRIVERS, separated mostly by their assists and turnovers. This is the cluster most similar to a traditional position: point guards.

OUTSIDE SHOOTERS

Code:
High: 3Att
Avg: AST, STL, qAST, USAGE
Low: OReb, 2Att, PF, DReb, BLK, FTA, HT, WT, TO
Bobby Phills, James Robinson, Glen Rice, Sean Elliott, Hubert Davis, Jim Jackson, Rasual Butler, Pat Garrity, Matt Bullard, Johnny Newman


The SHOOTERS. Defined mostly by their predilection for outside shooting, and by the low numbers in virtually every other stat category. Mostly smaller players, despite the presence of Garrity above:

Code:
PG 12%
G 19%
SG 28%
GF 21%
SF 9%
F 8%
PF 2%
FC 1%
C 1%


* * * * * * * * * * * * * * * * * * * * * *

The relationship between these clusters can be displayed on a 2-D "map", by plotting the first two discriminant functions. I love ascii graphics, so here you go:

Code:
+-------------------------------------------------------------+
|_)('.)('.)('.)('.)('.)('.)('.)('.)( |_____|_____|_____|_____||
|/ )('.)('.)('.)('.)('.)('.)('.)( ____|_____|_____|_____|___|
| \_\ )('.)('.)('.)('.)('.)('.)('.)( |_____|_____|_____|_____||
|__ _ )('.)('.)('.)('.)('.)('.)( ____|_____|_____|_____|___|
|/ \/ /('.)('.)('.'.'.'.'.'('.)('.)( |_____| ___ |_____|_____||
| \_\/ \)('.)('.'.DRIVING '.)('.)( ____|__.POST..__|_____|___|
|__ ___ )('.)( SWINGMEN )('.)('.)( |____.PLAYERS.____|_____||
|/ \/ / \/ )('.)'.'.'.'.'..)('.)( ____|_____ _____|_____|___|
| \_\/ \_\/('.)('.)('.)('.)('.)('.)( |_____|_____|_____|_____||
|__ ___ ___('.)('.)('.)('.)('.)( ____|_____|_____|_____|___|
|/ \/ / \/ / \)('.)('.)('.)('.)('.)( |_____|_____|_____|_____||
| \_\/ \_\/ \_\ )('.)('.)('.)('.)( ____|_____|_____|_____|___|
|__ ___ ___ __ )('.)('...'.)('.)(_|_____|_____|_____|_____||
|/ \/ / \/ / \/ /('.)('.).2.)('.)( ____ ___|_____|_____|___|
| \.\/ \.\/ \.\/ \)('.)('...'.)('.)(_|__ 1 |_____|_____|_____||
|_..BACKCOURT..___ )('.)('.)('.)(|_____ _____|_____|____|_.-.|
|/ BALLHANDLERS / \ )('.)('.)...)( |_____|___.-._'-._,-'_.-.|
| \_\/ \_\/ \_./ \_\)('.)('.)( .3.******|._,-'_.-._'-._,-'_.-.|
|__ ___ ___.6.__ __)('.)('.)-..*******._,-'_.-._'-._,-'_.-.|
|/ \/ / \/ / \. / \/ /***********.4:*****._,-'_.-._'-._,-'_.-.|
| \_\/ \_\/ \_\/ \_\/ ***********..******._,....-._'-._,-'_.-.|
|__ ___ ___ ___ _*****'*'.ROLE.******._,.5.;-._'-._,-'_.-.|
|/ \/ / \/ / \/ / \/ _****:PLAYERS.***-._,-._.-._'-._,-'_.-.|
| \_\/ \_\/ \_\/ \_\ 7. ***'''''''''***-._,-.;.;.;.;._,-'_.-.|
|__ ___ ___ ___ _.-**************-._,.DEFENSIVE.;-'_.-.|
|/ \/ / \/ / \/ /-._,-' *************'-._.SPECIALISTS.;'_.-.|
| \_\/ \_\/ \_\/ _.-.************'-._,..;.;.;.;.;,-'_.-.|
|__ ___ ___ '-._,-' ***********'-._,-'_.-._'-._,-'_.-.|
|/ \/ / \/ /.-._OUTSIDE_.-._***********'-._,-'_.-._'-._,-'_.-.|
| \_\/ \_\/ SHOOTERS '*********_'-._,-'_.-._'-._,-'_.-.|
|__ ___ _.-._ _.-._ ********_'-._,-'_.-._'-._,-'_.-.|
|/ \/ /_,-' '-._,-' '-********_'-._,-'_.-._'-._,-'_.-.|
| \_\/ _.-._ _.-._ ******._'-._,-'_.-._'-._,-'_.-.|
| '-._,-' '-._,-' '-._*****._'-._,-'_.-._'-._,-'_.-.|
| _ _.-._ _.-._ **'**._'-._,-'_.-._'-._,-'_.-.|
| '-._,-' '-._,-' '-._,****._'-._,-'_.-._'-._,-'_.-.|
|-._ _.-._ _.-._ **-._'-._,-'_.-._'-._,-'_.-.|
+-------------------------------------------------------------+

Symbol Label
------ --------------------

1 POST PLAYERS
2 DRIVING SWINGMEN
3 HUMAN CIGARS
4 MISC ROLE PLAYERS
5 DEFENSIVE SPECIALISTS
6 BACKCOURT BALLHANDLERS
7 OUTSIDE SHOOTERS
Group centroid displayed by group#


The numbers on the map show the cluster centroids. The territories for each cluster are well defined, except for the HUMAN CIGARS. That is because those players are not well defined themselves, except for the garbage time quality. You can see in the following chart how spread out they are:

[/img]

One last thing before I go. Here are some charts showing the relative diagnostic value of each stats used in determining cluster membership:






































_________________
ed
Back to top
View user's profile Send private message Send e-mail
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Sun Apr 09, 2006 7:44 pm Post subject: Cluster analysis Reply with quote
This is great Ed. Learning more about players compared to their peers by role is very important.


One definitional question:

AST: Percentage of teammate attempts assisted

Is that % of teammates attempts assisted by that player or assisted at all by anyone and is for all teammates and all time or just teammates on the floor concurrently with the player being studied?


Height and weight displays of each cluster would have value and then even more so various key performance metrics displayed at their physical location of the map to see if height and weight are positively correlated and how they are for different metrics and different clusters.

I’d also be curious to see % of total team time on court by these clusters, revealing team player type biases and weaknesses and then look at W-L records by these and note the patterns and think about how much meaning they have.

I wonder in how many cases key “misc. role players” actually counterbalance / address team minus them weaknesses. Is he the right role player for that team or just a role player with enough total quality points to contribute regardless of category and need. Most are forwards, I assume almost all teams have one in at least top 7 players but it would be interesting to note if any teams eliminate this type player and what type they substitute.

It is not surprising that in some cases who your teammates are and how strong they are on certain metrics can affect the clarity of your role cluster assignment. A two guard set evenly sharing the responsibilities might really share the ball handler and shooter cluster assignments and have lower than average ties to each. A PF/C combo with closer than normal relative post scoring ability might essentially share the post and defender roles which are usually divided. Shooter/ wing slasher can be shared as well.

Misc. role players may be misc. role players by nature or the other guys just may have taken most of the stats and left them that way, even though they could fill other roles if given that role opportunity.
Perhaps some of the best of the misc. role player lot may be undervalued.

Or the other way: The mid 90s Sonics definitely seemed to get a lot of value from Schrempf (went from Indiana where he was a little more of a post player to more strongly a point forward alongside Kemp) as the current Suns do Diaw, Horry on his various championship teams, etc. A misc. role player who can and does give you what you need game by game to win (shooting inside/outside, passing, rebounding, stops steals, etc.) is a very valuable thing.


A ball handler who can take and make threepointers at a good clip (above their cluster average) is quite valuable because of that if you want the threepoint game to play a larger than average role for the team. It is either that a higher volume 3 pt shooting forwards or both.

There must not be many average size guard defensive specialists as the cluster average height is over 6’ 8.5”. Ball handling and shooting needs usually trump? I assume of course some ball handlers and shooters are also good defensive players but just dont cluster as defensive specialist as their other attributes direct them more strongly to those clusters. A versatile 2/3, at least 6-7 is the main defensive specialist in the perimeter subset and some of the best like Bowen and Artest can cover 4 positions.


(With this work, as with Mike G.'s EWin work I would want to recommend use of a eFG% allowed and or points allowed number to address the missing one on one shot defense (and use of adjusted team def. rating on/off as a proxy for help defense) but I know many do not use the current 82games product because of concerns about quality and won't dwell on it. Some day it would be good if we all got over that hump satifactorily by full and careful use of video. Until then I'll use the best available data and homemade meta-ratings.)
Back to top
View user's profile Send private message
gabefarkas



Joined: 31 Dec 2004
Posts: 1291
Location: Durham, NC

PostPosted: Sun Apr 09, 2006 9:26 pm Post subject: Reply with quote
Ed, I heart you. This is truly phenomenal and thought-provoking.

My first instinct: instead of trying to entirely label players with only one category, do you think it would be possible to dole out "cluster credits" or something like that, where a player has a total of 100 points that are distributed by how they fit each criteria. For example, a player in the bottom center of your ascii graph (love it, btw), would probably be something like:

Outside Shooter = 27
Role Player = 20
Defensive Specialist = 25
Cigar = 5
Backcourt Handler = 12
Driving Swingman = 6
Post Player = 5

Does that make sense?
Back to top
View user's profile Send private message Send e-mail AIM Address
Ed Küpfer



Joined: 30 Dec 2004
Posts: 764
Location: Toronto

PostPosted: Sun Apr 09, 2006 9:29 pm Post subject: Re: Cluster analysis Reply with quote
Mark wrote:
One definitional question:

AST: Percentage of teammate attempts assisted

Is that % of teammates attempts assisted by that player or assisted at all by anyone and is for all teammates and all time or just teammates on the floor concurrently with the player being studied?


AST% = AST / ((TeamFGM - PlayerFGM)/(TeamMinutes/5))

IOW the proportion of teammates' made field goals (scaled to player's minutes) assisted by said player.


Quote:
Height and weight displays of each cluster would have value and then even more so various key performance metrics displayed at their physical location of the map to see if height and weight are positively correlated and how they are for different metrics and different clusters.


Among the mass of verbiage, I actually posted a chart of heights and weights by cluster. You must have missed it. Here's height:




Quote:
I wonder in how many cases key “misc. role players” actually counterbalance / address team minus them weaknesses.


One thing to keep in mind is the Role player cluster sits in the middle of the map, sharing "borders" with all the other clusters. This suggests to me that Role Players are drawn from every other cluster, presumably depending on the needs of the team, in addition to the changing abilities of the player. Just like a PG may shift over to the off-guard, or even the 3, based on the needs of the team (teammate injuries, matchups to be exploited, strategic surprise, etc) players of all types may move in and out of the Role Player cluster. This may or may not represent a change in the player's ability.

Keep in mind that the clusters are descriptive. The computer looked at a player's stats and said, oh, you were a role player last year, but this year you were an outside shooter. But once classified as a shooter, the player need not feel any impulse to remain in that role. The clusters were simply an after-the-fact description of what took place. If it turned out that winning teams had, say, more than the usual number of shooters, that does not suggest that teams should be looking to stock up on shooters. What it means is that winning teams tended to have players who played like shooters -- it doesn't mean those players were shooters.

I think the most interesting thing that can come of this is the study of player interactions at the game level. I don't think you can do too much by looking at teams at the season-level.

Quote:
Misc. role players may be misc. role players by nature or the other guys just may have taken most of the stats and left them that way, even though they could fill other roles if given that role opportunity.


Crap. You already said what I wrote.

Quote:
A ball handler who can take and make threepointers at a good clip (above their cluster average) is quite valuable because of that if you want the threepoint game to play a larger than average role for the team. It is either that a higher volume 3 pt shooting forwards or both.


I just want to make clear, so there's no confusion, that the stats I used, at least the shooting stats, were not efficiency stats, in that they don't say anything about how well a player shot, only how much he shot.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Ed Küpfer



Joined: 30 Dec 2004
Posts: 764
Location: Toronto

PostPosted: Sun Apr 09, 2006 9:43 pm Post subject: Reply with quote
gabefarkas wrote:
My first instinct: instead of trying to entirely label players with only one category, do you think it would be possible to dole out "cluster credits" or something like that, where a player has a total of 100 points that are distributed by how they fit each criteria.


There is, in fact, a clustering algorithm known as fuzzy k-means which does exactly what you're talking about, although it will reinvent its own clusters, which probably won't match the ones I have above.

In any case, before I get to the point where I can hand out cluster credits, I have to figure out a way to classify players into clusters based on their stats. What I did above was simply point out the existence of the clusters, but while I really like the clusters the computer discovered, the way the computer classified the players leaves much room for improvement. I'll probably end up going with a discriminant analysis approach, but a workable approach may also be to go the other way: just to classify players intuitively, based on the high-avg-low stats profiles I displayed above. I'm growing to like this approach -- it kinda makes sense, in that players are rarely classified into positions based solely on their stats. We don't look at a player' numbers and conclude that he is a small forward -- we use other information to classify his position. This is probably what we should do for clusters as well.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Mon Apr 10, 2006 1:20 am Post subject: Re: Cluster analysis Reply with quote
[quote="Ed Küpfer


"I actually posted a chart of heights and weights by cluster. You must have missed it. "

I read what you provided. I should have said a cluster specific height and weight chart with individual player names shown would be interesting in additional to your second chart of just dots (hard to react to that other than to say there is substantial variation) or the cluster average charts (which I appreciate as now you can say whether a player is bigger/smaller than role average and think about side by side with their production variances). General ones (not cluster specific) have been produced before I think by you and or Kevin P. so that itself isnt a big deal.

But as I said showing distibutions of names along with key performance metrics (central to that cluster's main role) still seems like it could have value especially if you got into FG% or TS% or certainly rebounding and some others where player size is a key input variable. You have summarized the averages for the clusters for many variables, I was just saying I would also have interest in the level of detail below that but of course then you are getting into many more charts. As much you care to share is welcome. And in addition to sharing it here I could see an article in the 82games/SI series or a sports journal if you care to publish anywhere else.



"I just want to make clear, so there's no confusion, that the stats I used, at least the shooting stats, were not efficiency stats, in that they don't say anything about how well a player shot, only how much he shot."

I know I was projecting beyond that but taking a quantity of three pointers and making a good percentage are both important. I was looking at the first characteristic on your chart but then would of course check the percentage made beyond what you have provided so far.

Thanks for your fine work and time responding.
Back to top
View user's profile Send private message
Ed Küpfer



Joined: 30 Dec 2004
Posts: 764
Location: Toronto

PostPosted: Mon Apr 10, 2006 12:32 pm Post subject: Reply with quote
I just want to add some thoughts to my first post here, about my intentions. What I wanted is to come up with an alternative to traditional positions, something that may prove more useful. The statistics used above are only there to confirm the existence of player clusters, but I could have just as easily come up with similar groupings just by thinking about it for awhile. For example, I tend to group players like Shaq, Bosh, Duncan into a category separate from Battie, the Collins twins, Mihm -- even though those players are traditionally categorised as centers or power forwards. In my head, I have a separate group for Kobe, Vince, Carmelo that I do for Mo Peterson, Adrian Griffin, Bruce Bowen. The small forward/off guard positional designations simply do not capture the diversity of these players. Isn't it more useful to think of the first group as scoring swingmen, and the second as perimeter defenders?

So I came up with some clusters, seven in fact, that I think represent natural player types. These aren't the only possible player types, but 7 is a reasonable number of categories to use. What I want is these clusters to be an alternative to traditional positions when analysing players. For example, ranking players within each clusters seems to me to be more reasonable than ranking players by position -- of course Vince is more productive than Bowen, but why are we comparing those two players? Shouldn't we compare Vince to other scorers, and Bowen to other defenders?

Different roles. It's important to keep these in mind. I don't want to be dogmatic about membership in each cluster. I spent a few hours yesterday trying to come up with methods of determining cluster membership for each player, but why should I? We don't spend much time looking at player stats to determine what position they play -- this is just something we know. That's how I want us to look at player clusters, as something we just naturally know. To that end, I'm not going to add to the confusion by posting a membership test. All you should really need to determine what cluster a player belongs to is your intuition (having a good prior knowledge of the definition of each cluster, of course). If you're still in doubt, use the player stats profiles I posted above.

POST PLAYERS
Code:
High: DReb, 2Att, FTA, WT, HT, USAGE, OReb, qAST, BLK
Avg: PF, TO, STL, AST
Low: 3Att


DRIVING SWINGMEN
Code:
High: 2Att, USAGE, FTA
Avg: STL, qAST, AST, 3Att, OReb, HT
Low: DReb, WT, PF, BLK, TO

ROLE PLAYERS -- PERIMETER DEFENDER
Code:
High: qAST
Avg: HT, OReb, WT, PF, DReb, STL, TO, BLK, 3Att, 2Att, FTA
Low: USAGE, AST


DEFENSIVE SPECIALISTS -- REBOUNDERS
Code:
High: WT, HT, BLK, DReb, OReb, PF, qAST
Avg: TO, FTA
Low: USAGE, 3Att, AST, STL, 2Att

BACKCOURT BALLHANDLERS -- DISTRIBUTORS
Code:
High: AST, STL, TO
Avg: 3Att, USAGE, FTA, 2Att
Low: qAST, HT, WT, OReb, DReb, BLK, PF

OUTSIDE SHOOTERS
Code:
High: 3Att
Avg: AST, STL, qAST, USAGE
Low: OReb, 2Att, PF, DReb, BLK, FTA, HT, WT, TO

HUMAN VICTORY CIGARS
Code:
High: PF, TO, STL, FTA, USAGE
Avg: OReb, qAST, 2Att, 3Att, HT, WT, DReb, BLK
Low: AST


These clusters came from statistical analysis. But there's no reason the definitions have to remain static. Where, for example, are the perimeter defenders? The stats for these players don't really capture the nature of their ability, unfortunately, so the cluster analysis didn't "find" them. I believe they would be split among the OUTSIDE SHOOTER and ROLE PLAYER clusters. But perimeter defenders are easily conceptualized, even if the stats don't see them. I'm going to change the ROLE PLAYER cluster to include them. I'm also going to take a suggestion from DeanO and rename the DEFENSIVE SPECIALISTS cluster to REBOUNDERS, and the BALLHANDLERS to DISTRIBUTORS.

The last thing is the CIGARS. This is, almost by definition, a garbage can cluster, comprising mostly of players who don't fit into the other clusters. Essentially, there are six clusters, plus one for the players who don't get much playing time. I think in most analysis we can ignore the CIGARS, which would mean 6 categories of players.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Ed Küpfer



Joined: 30 Dec 2004
Posts: 764
Location: Toronto

PostPosted: Mon Apr 10, 2006 11:19 pm Post subject: Reply with quote
Just in case someone is dying to have an automatic way of classifying these players, here's a relatively painless method. To calculcate the probablity for each player belonging to any particular cluster, you'll need the following:

Code:
HEIGHT = Player height in inches minus 60
2ATT = 2-point attempts per minute
3ATT = 3-point attempts per minute
FTA = Free throw attempts per minute
OR = offensive rebounds per minute
DR = defensive rebounds per minute
TO = turnovers per minute
AST = assists per minute
BLK = blocks per minute


The probability of a player belonging to a cluster is calculated by the logistic equation

probablity of belonging to cluster C = 1/(1 + EXP (-(B)))

The variable coefficients for each player cluster follow.

Code:
CONSTANT HEIGHT 2ATT 3ATT FTA OR DR TO AST BLK
CIGAR -2.9 -0.035 -4.9 5.2 8.3 9.5 -9.0 36.2 -15.4 -23.4
DISTRIBUTOR 4.1 -0.514 -8.3 -9.6 -4.4 -17.1 -1.4 6.3 52.3 -19.9
DRIVER -4.6 -0.005 21.2 -19.4 7.2 -14.6 -22.2 -8.1 -9.8 -34.9
PERIMETERD 2.5 0.075 -4.7 -13.4 -7.9 -1.1 -6.2 -12.6 -10.2 -34.3
POST -16.8 0.324 14.6 -21.4 12.9 2.6 18.2 -6.1 -2.5 -8.0
REBOUNDER -6.3 0.450 -17.7 -27.3 -15.6 17.7 13.8 3.0 -27.8 25.1
SHOOTER 1.1 0.157 -4.2 45.0 -8.5 -36.9 -15.0 -36.2 -15.7 -24.8


For example, here is Ilgauskas (2005):

Code:
PLAYER HEIGHT 2ATT 3ATT FTA OR DR TO AST BLK
Z Ilgauskas 27 0.374 0.003 0.192 0.114 0.143 0.073 0.038 0.063


To calculate his probablity of beloging to the CIGARS cluster, multiply each of those numbers by the CIGARS coeficients from the table above, and then add them all together (including the constant). I get -3.73 as my sum.This sum is the B that goes into the logistic equation:

probablity of belonging to CIGARS cluster = 1/(1 + EXP (-(-3.73))) = 2%

Now go through that procedure for each of the clusters.

Code:

CLUSTER B p(Cluster)

CIGAR -3.73 2%
DISTRIBUTOR -14.70 0%
DRIVER -3.46 3%
PERIMETERD -3.27 4%
POST 1.62 84%
REBOUNDER 0.81 69%
SHOOTER -8.97 0%


And there you go. That's the easy method — trust me, you don't want to see the complicated way. This way does an excellent job (80%+ accuracy) in identifying all cluster except CIGARS, where it doesn't do a good job at all. But that's okay. The other cluster that maybe needs some improvement is PERIMETER-D, where my short method has an accuracy of only 60%. This is the cluster that most requires subjective judgment on the part of the observer to identify, since most of the things these players do well don't show up as stats.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Ben F.



Joined: 07 Mar 2005
Posts: 391


PostPosted: Tue Apr 11, 2006 3:20 am Post subject: Reply with quote
Ed,

You say that you hope these role designations are more "useful". Does that mean you are trying to come up with a better description of what the players actually do, or that you are trying to come up with something that could be used to help a team?

If it's the latter, I'd like to see how you could use this analysis (which I think is incredible, by the way) to answer what I call the "Boris Diaw Dilemma". (Maybe this should be a separate thread if it can't be answered by this analysis.)

The prevailing theory of Boris Diaw's incredible improvement this year is not just simple work ethic - it's a matter of roles. The Hawks obviously did not know how to use Diaw effectively. They had him at PG (or "distributor") often. Now, D'Antoni has him in the forward and center positions, and it's entirely changed his game. So the question is twofold:

1) How can we identify players who are used in the wrong roles?
2) How can we identify what role they'd be most effective in?

The idea that Diaw can make this incredible leap makes me wonder how many diamonds in the rough there are out there, who are just being misused - especially in this era of players that can do everything.
Back to top
View user's profile Send private message
jeffpotts77



Joined: 18 Feb 2005
Posts: 150
Location: Cambridge, MA

PostPosted: Tue Apr 11, 2006 10:30 am Post subject: Reply with quote
I have nothing to add except to say that this is brilliant! Thanks for you hard work, Ed! Your point about comparing Vince Carter to Bruce Bowen is spot-on. Will you be publishing this anywhere else for non-apbr members to see?
Back to top
View user's profile Send private message Yahoo Messenger
Ed Küpfer



Joined: 30 Dec 2004
Posts: 764
Location: Toronto

PostPosted: Tue Apr 11, 2006 1:55 pm Post subject: Reply with quote
Using the method outlined above, I classified the players for three different 2006 teams. Each player is has a CLUSTER membership, which is simply the identity of the cluster with the highest probability. Those probabilities are shown in the 6 rightmost columns, as percentages. For example:

Code:
PLAYER TEAM PS MPG CLUSTER DIST DRIV PERI POST REB SHOOT
Boris Diaw PHO SF 35 DISTRIBUTOR 19 0 6 2 1 0
Eddie House PHO SG 17 SHOOTER 12 0 2 0 0 100


Diaw is classified as a DISTRIBUTOR, because of his 19% probability of belonging to that cluster. But 19% is fairly low -- it would be more accurate to say that none of the clusters captures his stats profile very well at all (in his two previous seasons, he was classified as a PERIMETER-D and DISTRIBUTOR, again with very low probability scores). Eddie House on the other hand is classified unambiguously as an OUTSIDE SHOOTER, with a small nod to DISTRIBUTOR. I think this matches up with reality fairly closely.


Code:
PLAYER CLUSTER DIST DRIV PERI POST REB SHOOT

Boris Diaw DISTRIBUTOR 19 0 6 2 1 0
Eddie House SHOOTER 12 0 2 0 0 100
James Jones SHOOTER 0 0 15 0 2 100
Kurt Thomas REBOUNDER 0 0 19 17 81 0
Leandrinho Barbosa SHOOTER 10 1 11 0 0 84
Pat Burke REBOUNDER 0 2 9 16 33 0
Raja Bell SHOOTER 1 0 24 0 0 100
Shawn Marion REBOUNDER 0 0 4 4 5 4
Steve Nash DISTRIBUTOR 100 0 1 0 0 3
Tim Thomas SHOOTER 0 0 10 0 1 97


Code:
PLAYER CLUSTER DIST DRIV PERI POST REB SHOOT

Beno Udrih DISTRIBUTOR 57 1 4 0 0 46
Brent Barry SHOOTER 3 0 12 0 1 98
Bruce Bowen SHOOTER 0 0 56 0 12 97
Emmanuel Ginobili SHOOTER 5 1 2 0 0 34
Fabricio Oberto REBOUNDER 0 0 43 1 92 0
Michael Finley SHOOTER 0 0 26 0 0 99
Nazr Mohammed REBOUNDER 0 0 17 19 89 0
Nick Van Exel SHOOTER 62 0 8 0 0 97
Rasho Nesterovic REBOUNDER 0 0 42 2 99 2
Robert Horry SHOOTER 0 0 12 0 28 77
Tim Duncan POST 0 1 1 90 4 0
Tony Parker DISTRIBUTOR 86 55 5 1 0 0


Code:
PLAYER CLUSTER DIST DRIV PERI POST REB SHOOT

Andre Barrett DISTRIBUTOR 100 6 8 0 0 2
Antonio Davis REBOUNDER 0 1 57 2 80 1
Charlie Villanueva SHOOTER 0 0 13 2 10 22
Chris Bosh POST 0 15 6 78 2 0
Darrick Martin DISTRIBUTOR 99 0 11 0 0 76
Eric Williams PERIMETER-D 0 0 45 0 7 42
Joey Graham PERIMETER-D 0 1 36 0 7 23
Jose Calderon DISTRIBUTOR 99 0 19 0 0 2
Loren Woods REBOUNDER 0 0 23 4 100 0
Matt Bonner SHOOTER 0 0 33 0 22 99
Mike James DISTRIBUTOR 76 1 3 0 0 57
Morris Peterson SHOOTER 0 0 17 0 0 98
Pape Sow REBOUNDER 0 0 37 3 95 0
Rafael Araujo REBOUNDER 0 0 50 5 91 0



This method doesn't like to find POST players -- it only classified 20 players that way in 2006. By tweaking the constant in the equation for POST, you can increase this number, or you can simply choose to see POST players as ELITE POST players, and 20 as a reasonable number.

Toronto is the team I know best, and looking at the results, I am very happy with the classificiations. Villanueva's game is hard to classify, and so I'm content with the low scores he registered as a SHOOTER, PERIMETER DEFENDER, and REBOUNDER. Bosh is of course an elite POST player, and Calderon is a DISTRIBTOR in the old-school PG style. Mike James gets high scores as a DISTRIBUTOR and SHOOTER, which is just right.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
94by50



Joined: 01 Jan 2006
Posts: 499
Location: Phoenix

PostPosted: Tue Apr 11, 2006 3:17 pm Post subject: Reply with quote
How useful would all this knowledge be in developing a method for judging how similar two players are? Strict statistical similarity scores are helpful, but perhaps this could be another step in that direction.
Back to top
View user's profile Send private message
THWilson



Joined: 19 Jul 2005
Posts: 164
Location: phoenix

PostPosted: Tue Apr 11, 2006 4:12 pm Post subject: Reply with quote
Ed Küpfer wrote:

This method doesn't like to find POST players -- it only classified 20 players that way in 2006. By tweaking the constant in the equation for POST, you can increase this number, or you can simply choose to see POST players as ELITE POST players, and 20 as a reasonable number.


It also seems to have trouble with Drivers. These are the two high-usage groups, and usage isn't in the linear weights...any connection? I was really surprised to see Manu only get a 1 for driver...

Commendable work, btw.
Back to top
View user's profile Send private message
ziller



Joined: 30 Jun 2005
Posts: 126
Location: Sac Metro

PostPosted: Tue Apr 11, 2006 5:20 pm Post subject: Reply with quote
I took the liberty of running another team through Ed's magnificent gauntlet.



(I can't for the life of me get columns lined up properly. So you get an image. Sorry.)

The various "perimeter-d" cluster members worry me - it makes more sense in this instance for the original role-player tag. The only other problem would be the low marks for "driver" for Kevin Martin and Bonzi Wells. Martin gets to the line plenty, so that's not hurting him. Perhaps here it's a relative lack of two-point field goals - Martin is either taking a three or slashing, a la Joe Johnson 2005. Bonzi doesn't get to the line consistently, however, so that could be his low driver reason.
_________________
SactownRoyalty.com
tziller@gmail.com
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Tue Apr 11, 2006 5:34 pm Post subject: Reply with quote
The point distribution detail is interesting (almost essential to see really to me).

"There are 6 clusters". OK if you are already set on that based on what the analysis is telling you. You've indicated in the post that follows you essentially dont want a hybrid system (even though the clusters are a mix of role and physical attributes (arent they pretty powerful membership tests?). Wouldnt a pure "role system" remove height and weight? And the cluster then just be distribute, outside shoot, drive, defend, rebound, post without qualifiers by size or location on the court? Have you run the data to produce clusters without height and weight? I assume it is messier to look at -compared to traditional position classification- but isn't that good? I recall in other threads you wrestled with these questions so would welcome any further explanations you care to add about your thinking on this issue now.

I'll still use some form of position/role dual way of thinking about players. Position isnt enough alone, case closed. Role using 6 clusters (found with weight/ height) certainly is valuable. It still seems like a hybrid system to me and distance from the cluster dividing line seems to carry a lot of importance and that suggests the main cluster label isnt doing enough alone. A system with more clusters is a reach for a unified system. Wouldn't they be tighter to the cluster means that way? I started some discussion of options for a 8,9 or 12 cluster hybrid direction but if that moves too far from what you are doing I will hold off on that here and now.

Last edited by Mark on Wed Apr 12, 2006 10:43 am; edited 8 times in total

Author Message
Ed Küpfer



Joined: 30 Dec 2004
Posts: 787
Location: Toronto

PostPosted: Tue Apr 11, 2006 5:56 pm Post subject: Reply with quote
94by50 and THWilson:

I don't want to get ahead of myself here. My primary goal is to establish these clusters as groupings with just as much reality and utility as the PG-->C spectrum. I want people to look at Richard Jefferson and catalogue him as "driver" the same instinctive way they think of him as a small forward. I believe these clusters represent "territories" of suites of player abilities. Some players live in the centre of a territory (maybe the capitol city, to extend the metaphor), others live near the border of adjacent territories. Since these territories are not confined to the usual two dimensions, maybe some players live at the intersection of three or more territories.

To the extent that these clusters, the territories, are real, I want everyone to remember that the statistical profile of each cluster is merely an approximation of the abilities that characterize that particular cluster. Think of shooting ability: it is a real skill, a combination of many physical and mental factors. We estimate shooting ability through some shooting percentage, but that shooting percentage is not the actual thing -- it is only an approximation of the actual skill. In the same way, the methods I posted above are not actual descriptions of the clusters -- they are only approximations, and in one or two cases, not particularly accurate approximations. But the clusters themselves are real: Manu is a slashing driving swingman, no matter what my approximations say. Keep that in mind.

Which goes to Ziller's post. (Nice job, by the way.) If there's some confusion about the PERIMETER DEFENDER/ROLE PLAYER cluster, it's because we've only begun defining the cluster itself. So far, I've only noted that some statistical methods tend to group a certain type of player that we might call role players or perimeter defenders. But the title of the cluster is not the cluster itself: the group of players is primary, and if we have to change the title to better describe the group, then that's what we should do. (I don't want to change the name of the group yet. I just wanted to point out that I believe in the existence of these clusters, but I'm not quite sure if we can define them precisely.)

The take home message is this: there are six clusters. Think of any player and try to put him into one of those clusters. If he doesn't fit easily into any, that's okay. We can't always agree on whether some players are power forwards or centers. There's no reason why every player needs to be only in one cluster, or why any player's game can be adequately described by any cluster. I just want these clusters to change the context in which we see player roles and abilities. I don't want to talk about "pure" point guards and "scoring" point guards any more -- I want to talk about distributors and drivers.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
ChadC



Joined: 25 Jul 2005
Posts: 12
Location: Illinois

PostPosted: Wed Apr 12, 2006 12:09 am Post subject: Reply with quote
I don't have much content to add, but I wanted to give a well-deserved "Bravo!" to Ed. While this is still in developing stages, it is one of the most exciting things I've read here in a long time.

Echoing a previous post, I'm also wondering about the possibility of using these categories to see if using a player in another role might be better (a la Boris Diaw).

Other ways to use this idea may be to look at compositions of successful teams vs. the bottom feeders. Is it better for players to have well defined roles or for them to be versitile?

Or look at how players evolve over their careers?

Or maybe look at how a coaching style produces certain players? How much does the coach affect the team's play style and how much is affected by the players on the team?
_________________
-Chad Casarotto
Back to top
View user's profile Send private message Send e-mail AIM Address
Ed Küpfer



Joined: 30 Dec 2004
Posts: 787
Location: Toronto

PostPosted: Wed Apr 12, 2006 2:14 pm Post subject: Reply with quote
(This post is only marginally related to the topic. Feel free to skip it.)

If we posit varying degrees of cluster membership based on some test, ranging from 0 = 0% chance of belonging to a particular cluster to 1 = 100% chance of belonging to a particular cluster, there will be some confusion when it comes to players who A) are not unambiguously members of a cluster (ie low cluster probability score), and B) are unabiguous members of more that one cluster (ie have high cluster probability scores for more than one cluster). These don't pose real problems for the idea of players clusters, since nothing prevents a player from having wide-ranging non-specialised abilities, but it does pose problems for us in classifying the players. I'm going to propose that cluster membership be augmented with two further measures: Sensitivity and Positive Predictive Value.

In binary classifications, sensitivity is calculated as the the ratio of true positives to the sum of true positives and false negatives. In our case, the true positive is the maximum probability score of the 7 clusters. The false negative is the compliment of the maximum probability score. Sensitivity boils down to the highest probability score among all clusters.

Positive Predictive Value is defined as the ratio of true positives to all positives, or how clear the answer of cluster membership is -- as PPV approaches 1, the clearer the answer. As PPV approaches zero, the more confusion. To calculate, take the maximum score and divide by the sum of all scores.

Take two cases.

Code:
CIGAR DIST DRIVE PERI POST REB SHOOT SENSITIVIY PPV
100 99 90 0 0 0 0 100 35
100 5 2 2 0 0 0 100 92


All numbers are percentages. In the first case, the sensitivity is 100%. This player scored highly on his cluster membership test. On the other hand, his PPV is only 35% (PPV = 100 / (100 + 99 + 90), which means there is confusion about exactly which cluster he is a member of. The second case is clearer: sensitiviy also equals 100%, but his PPV is 92%, which means we can be more sure he belongs to one, and only one, cluster.

Here are two more cases:

Code:
CIGAR DIST DRIVE PERI POST REB SHOOT SENSITIVIY PPV
32 0 0 0 0 0 0 32 100
32 24 13 10 3 0 0 32 39


Here the sensitivity is much lower. The player had a high membership score of only 32%. But with a PPV of 100%, we can be more sure that to the extent that he belongs to any one cluster, he belongs only to that cluster. The next case also has a sensitivity of only 32%, but with other similar scores in other clusters, his PPV is only 39%, meaning that we can't be sure of his cluster membership.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Ed Küpfer



Joined: 30 Dec 2004
Posts: 787
Location: Toronto

PostPosted: Wed Apr 12, 2006 8:24 pm Post subject: Reply with quote
Cluster membership trends.




(Click the image to see enlarged version.)

The thing that really jumps out is the decline of the DRIVER, mirrored exactly by the rise of the SHOOTER. This is somehow realated to three-point strategy, but I'm not quite sure how right now. Membership in both clusters has remained stable for 10 years, suggesting something or other. (I'm not at my analytical best right now.) Notable also is the gradual rise of the REBOUNDER, but perhaps most interesting (to me anyway) is the nearly stable membership in POST PLAYERS, since I'm always hearing about how these types of players have largely disappeared from the landscape.

I forgot to mention, the chart shows three-year moving averages, just to smooth everything out a little.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Ed Küpfer



Joined: 30 Dec 2004
Posts: 787
Location: Toronto

PostPosted: Wed Apr 12, 2006 8:52 pm Post subject: Reply with quote
I looked at All-League seasons to see which clusters were represented:

Code:
Lg DISTR DRIVER PERI POST REB SHOOTER

nonAll-Lg 17 16 16 13 21 13
All-Lg 18 28 1 44 4 5


All numbers are percentages. The table should be read as "17% of non-All-League players were classified as Distributors, 28% of All-Leaguers were Drivers."

Here's how they break down if you split the All-Leaguers into 1st Team, 2nd Team, and 3rd Team:

Code:
Lg DISTR DRIVER PERI POST REB SHOOTER

1st Team 17 28 1 51 1 1
2nd Team 20 31 1 40 3 4
3rd Team 17 23 2 36 10 12


Looks like Perimeter defenders/Role players are not thought of very highly. And while voters are reluctant to honour Rebounders and Shooters as 1st and 2nd teamers, those players often get the nod for 3rd team.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 12, 2006 10:17 pm Post subject: Reply with quote
Since 2pt shot attempts is a key factor separating drivers from shooters, I wonder have the cluster averages for this stat changed over time?

Does historical data exist on charges called and is it a rising curve and enough to be consider a significant factor in the decline of drivers? The general rise in emphasis on defensive effort and pseudo zone defenses ?

Could the rise of salary increase the competition among players to score and get paid and lead to fewer drivers with high 2pt attempts per min and more players with lower attempts per min.?

Could the rise of salary levels also offer a bigger explanation for avoiding a style of play- driving- that might expose one more to injury/loss of income, with the three point shoot providing a strong reason on its own but a convenient alternative for players concerned by this?

Any difference in distribution between the two clusters for young players vs older vets?
Back to top
View user's profile Send private message
bchaikin



Joined: 27 Jan 2005
Posts: 690
Location: cleveland, ohio

PostPosted: Fri Apr 14, 2006 8:52 pm Post subject: Reply with quote
The prevailing theory of Boris Diaw's incredible improvement this year is not just simple work ethic - it's a matter of roles. The Hawks obviously did not know how to use Diaw effectively. They had him at PG (or "distributor") often. Now, D'Antoni has him in the forward and center positions, and it's entirely changed his game....The idea that Diaw can make this incredible leap ...

a few points on mr. diaw:

this 05-06 season in phoenix his touches/min are 25% more than what they were in atlanta in 04-05. but they were 25% more in atl in 04-05 than what they were in atl in 03-04. these are large increases in consecutive seasons, but early in a player's career is typically when you see the most increase in their touches (their first few seasons, especially their 2nd to 3rd and 3rd to 4th seasons in the league)...

this season he is shooting the ball with 1/5 of all his touches and passing it with 2/3 of his touches, virtually identical to what he did in 04-05 and 03-04. these are attributes typically associated with a PG, and other players this season with similar touches/min and similar percentages of what they do per touch include t.j. ford, rafer alston, anthony johnson, and earl watson. a frontcourt comparison with a similar "game" would be lamar odom, but his touches/min are 15% less than diaw's...

he is getting fouled on 1/20 of all his touches, same as in 04-05 and 03-04...

his rebounding this season per 48 min is only 7%-8% more than what he got in 03-04, and is less than what your average PF is getting this season but more than what your average SF is getting...

the big differences i see from his earlier seasons have been (a) his turnovers per touch this season are 1/2 of what they were during his rookie campaign two seasons ago, (b) his 2pt shooting is up almost 7% from his previous best (54% vs. 47%), and (c) his defense outside of steals, blocks (these are also up), and defensive rebounding is very good...

so his touches on offense are up from his atlanta seasons but he's doing pretty much per touch what he did in each of his previous hawks seasons, but he's shooting better and turning the ball over less. i'd say his role is the same in phoenix as it was in atlanta, and that when you look closely at his stats they show that his game - other than his better shooting and reduced turnovers, typical when a player enters his 3rd and 4th seasons - really hasn't changed much at all. i don't believe it had anything to do with the hawks not knowing how to use him, but more the fact that he was just 21 and 22 years of age when with the hawks, didn't shoot particularly well in atlanta (like most rookies and young 2nd year players don't), and they gave him up to get a better shooter, joe johnson, who had shot almost 48% on close to 400 3pt attempts, and whom they had thoughts of converting into a PG. throwing in the two 1st round draft choices was the insane part of the deal imho...

now diaw's a frontcourt "PG/distributor" rather than a backcourt one, similar to how vlade divac was in sacramento two seasons ago. actually their stats (diaw in 05-06 and divac in 03-04) are similar in a number of ways - their touches/min, what they did per touch, rebounding, steals, and blocks, even how often they committed fouls are about the same...

sounds more to me like d'antoni took a backcourt passer who could rebound and play D and simply switched him to the frontcourt...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Doc319
Guest






PostPosted: Sat Apr 15, 2006 6:37 pm Post subject: Reply with quote
I am intrigued by the idea that Diaw is actually doing the same things, per touch, that he was in Atlanta. I have not heard anyone else suggest that. I have heard it suggested that he was always capable of doing these things or that he did them when he played for the French National Team, but that is different than saying he actually did these things in Atlanta.

Of course, after he arrives in Phoenix and is successful playing for D'Antoni it is easy to say that D'Antoni "took a backcourt passer who could rebound and play D and simply switched him to the frontcourt..." That's like "predicting" on Saturday that Kobe will score 50 points against Portland on Friday. We all see how D'Antoni has used Diaw. The question is, could simulation have predicted this result before it happened? When Phoenix made this trade, Diaw was basically a throw-in. Did simulation predict before the season that Diaw would put up the kind of numbers that he has this year? I am curious which players, according to simulation, are currently being underutilized by their coaches and would blossom if they played more extensive minutes. Who are some candidates to be next year's Boris Diaw? Obviously, this is dependent on their coaches utilizing them properly but if simulation can identify several underutilized players then we can track them over a period of time and follow their productivity.

--David Friedman
http://20secondtimeout.blogspot.com/
Back to top

bchaikin



Joined: 27 Jan 2005
Posts: 690
Location: cleveland, ohio

PostPosted: Sat Apr 15, 2006 9:44 pm Post subject: Reply with quote
I am intrigued by the idea that Diaw is actually doing the same things, per touch, that he was in Atlanta. I have not heard anyone else suggest that.... Of course, after he arrives in Phoenix and is successful playing for D'Antoni it is easy to say that D'Antoni "took a backcourt passer who could rebound and play D and simply switched him to the frontcourt..."

it has nothing to do with whether he was successful or not, the simple fact is he is now doing per touch what he had done his previous two seasons in atlanta, in terms of how often he shoots, passes, and gets fouled per touch (he's cut his turnovers down alot). the key differences are another large increase in his touches/min, better shooting, and that reduction in turnovers (better shot blocking too)...

but in terms of his "game", per touch he's doing pretty much the same now as he did in atlanta - both seasons. even his rebounding now in phoenix is pretty much the same as his rookie season as a hawk (less than 10% better now), despite playing all 3 frontcourt positions now versus guard and some small forward as a hawk rookie...

That's like "predicting" on Saturday that Kobe will score 50 points against Portland on Friday.

actually its more like saying if anyone on friday in the lal/por game was going to score 50, odds are it would have been bryant...

We all see how D'Antoni has used Diaw.

how has d'antoni used diaw? other than switch his position(s) played?...

The question is, could simulation have predicted this result before it happened?

this is simply stats analysis... looking at the numbers and drawing conclusions from them. the simulation does not determine a player's touches/min nor what he does per touch - the player does, with some influence from his teammates...

the nba stats database at www.bballsports.com is free to the internet public, has been for quite some time, and all player touches/min and what they did/do per touch are listed there. you can see diaw's numbers there...

I have heard it suggested that he was always capable of doing these things or that he did them when he played for the French National Team, but that is different than saying he actually did these things in Atlanta.

he actually did these things in atlanta - how often he shot, passed, and got fouled per touch, just like he is now in phoenix, but with significantly more turnovers per touch (typical of a rookie), and less touches/min...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Sun Apr 16, 2006 12:28 am Post subject: Consideration of players in other roles, etc. Reply with quote
Diaw before this year at b-r.com had a high similar score with most notably Andre Iguodala but he is a wingman. Would still be interesting to see his role split side by side to Diaw. Prior to thius year Iguodala had a slightly higher rebound and only a modestly lower assist rate while scoring more.
http://www.basketball-reference.com/lab ... it=Compare
This year Diaw is ahead on all three, modestly on points and rebounds more pronounced on assists.

Josh Childress is listed as one of the most similar to Iguodala and remains more similar to him than Diaw though could he be a little more able to become like Diaw is this year in the future if he puts on some weight and Harrington leaves? Seems pretty strong and I had thought he could dribble & pass above average coming into the league.

But both probably are stretches too far away from just being driving wings. Any better choices for possible future similarity to the new Diaw?

Luol Deng seems a better comparison to Diaw physically and perhaps skills wise on interior defense and passing. He played almost 10% of his time this year according to 82games at PF and the matchup results are pretty favorable. Pretty good assist rate. Maybe more PF play next year?

Ike Diogu could improve too and center by the numbers is his better position so far.

I like DerMarr Johnson's height and think he is pretty versatile perhaps with some Diaw breakout potential on right team in a featured role. He might be able to play PF on some teams. His one on one and team defense there is a few looks there are good this year and he can spread the floor and hit the three. 5% of his time this year was at PF and it was his best position matchup results. If I wanted a mobile shooting forward who could create mismatches at PF occasionally I might look at him as that but he gets outrebounded badly at PF so you'd have to be really good at center and elsewhere too.


Using Ed K's earlier similarity database other Diaw comparison candidates include Jalen Rose (did well at PF in 3% of his time in Toronto- enough offense, good defense and rebounding this year, and did that well last year in about 3% of time too except for the rebounding part) and perhaps more based on last year Bobby Simmons is a sturdy player capable of playing some PF.

Pietrus surprisingly he does score better at PF than anywhere else this year, his defense is very good there this year in 6% of his time but he does get badly outrebounded and doesnt pass alot yet. But another intriguing hard to classify yet guy like Diaw was. And French too.

Rooting around you could also mention as similiar to some of Diaw's similars Darius Miles, Turkoglu, perhaps Caron Butler and Battier. None pass like Diaw when playing interior but are decent teammates with the ball and they can defend at least PF part of the time ok. Damien Wilkins has a similar body type and did pretty well at PF last year and is developing as a passer this year. James Posey in Memphis had a little time at PF and did ok last year.

Luke Walton? Defends, rebounds, passes very well from inside. 12pts 9 rebs 6 assists per 48. If you were looking for the next Diaw on all skills he might fit. Dont know if he was further upside but I might encourage an effort to get him from the Lakers.

Kukoc can defend PF and pass well and might be one of the closest to Diaw this year Kukoc 15pts 7 rebs 7 assists per 48 ; Diaw 18 pts 9 rebs 8 assists. younger Toni even more so- 19 pts 8.5 rebs 6 assists in 2003-4, 20.5 pts 6.5 rebs 6 assists on 2002-3 per 48. And of course Odom.

Lebron James has played 2-3% of his time at PF and he destroys them by 40 points per 48 minutes!! And wins the rebounding matchup and continues to pass well.


Other players who might be better at other positions and/or roles:

If Boston didnt have Pierce and Wally, Ryan Gomes might do better as a wing than PF were he mostly plays right now and is doing alright but he might dominate as a wing?

Anderson Varejao mostly plays center, Dougstats mistakenly lists him as a SF but I was already wondering perhaps if he could be Kirilenko like as a forward? His defense is much better at PF than C but Cleveland has little room there this year. Might change is future if Gooden isnt retained. (Pavlovic is much better at SF but James is there.)

Marquis Daniels is / has been better at PG but Dallas filled that time with two others. He is not bad as a wing but he seems more like a driving, defending, decent passing PG to me. I could see him changing teams to get more playing time more money and possibly the PG job if he cares to be there.

B Davis seems miscast as #1 scorer. I doubt you could make him less a shooter now (except maybe on a champioship caliber team with a coach he got along with and emphasis his good defense even more) but that might be the better mix for the team's sake.

Could K Bogans play more like Billups? He show some ball handling/playmaking in college but isnt likely to be that in NBA without a huge change of thinking/playing (but being a primary ballhandler and getting more minutes/shots might be enough to appeal to him do pass some too with the right team/coach and slowly grow into it? A longshot but could boost his value from marginal now). Should be more a driver not shooter.

Last edited by Mark on Sun Apr 16, 2006 5:19 pm; edited 23 times in total
Back to top
View user's profile Send private message
Doc319
Guest






PostPosted: Sun Apr 16, 2006 1:12 am Post subject: Reply with quote
Quote:
how has d'antoni used diaw? other than switch his position(s) played?...


By "used" I mean that D'Antoni has increased his minutes and given him significant playmaking responsibilities (second on the team to Nash in assists). Whether or not he was doing the same things per touch in Atlanta, he had a lot fewer minutes and touches when he played there. Now he is a significant player on a division champion team after being a bit player on a lousy team.

Would you ascribe more significance to the touches of a 35 min. a game player who likely is playing a lot against the other team's starters versus the touches of a 25 min. a game player who likely played a lot against reserves? Or does the sim's programming assume that such considerations are not important over the course of a season? What I am getting at is that, granting for the sake of discussion that Diaw continues to basically do the same things per touch, is it significant that in one situation he played less minutes (i.e. would be more rested, less apt to be concerned about foul trouble that could curb his aggressiveness, etc.) and probably played against bench players and that in the current situation he plays more minutes and probably plays more against the top players? An extreme example that those of us who play pickup ball will understand: player "A" may be a big scorer or rebounder at certain playgrounds/rec centers, while at other playgrounds/rec centers he may hardly score or rebound at all, if he can even get on the court. His "game" is the same but what he is able to do is affected by the level of competition. Isn't there some signficance to Diaw maintaining the same scoring/passing tendencies despite playing more minutes against (presumably) tougher competition?

I understand that on a per minute basis, according to the stats that the sim tracks, Diaw is doing similar things per touch (other than decreasing his turnovers and increasing his shotblocking, as you noted). For whatever reason, Atlanta either didn't see much value in what he was doing or thought that the players ahead of him in the rotation contributed more value (perhaps such evaluations are why they miss the playoffs year after year). Are there other Diaws out there, according to the sim? Are there other players playing 25 min./g or less, like Diaw did in Atl., who have shoot/pass per touch stats that suggest that they would flourish if their minutes and responsibilities were increased? I don't think that anyone, including Phoenix, anticipated Diaw's level of production this year. Would you support Diaw's candidacy for Most Improved Player or, since your sim suggests that he is playing the same "game," would you vote for someone else?

--David Friedman
http://20secondtimeout.blogspot.com/
Back to top

bchaikin



Joined: 27 Jan 2005
Posts: 690
Location: cleveland, ohio

PostPosted: Mon Apr 17, 2006 12:12 am Post subject: Reply with quote
By "used" I mean that D'Antoni has increased his minutes and given him significant playmaking responsibilities (second on the team to Nash in assists).

d'antoni may have increased diaw's minutes but he didn't give him significant playmaking responsibilities, anymore than d'antoni gave steve nash playmaking responsibilities. the suns signed nash for what he was - a player that was an excellent offensive player that was also a very good passer. they also signed (rather asked for in trade) diaw because he was a good passer that could rebound and play D...

saying d'antoni gave diaw or nash playmaking responsibilities is like saying the suns traded for shaq and then gave him rebounding and scoring responsibilities...

Whether or not he was doing the same things per touch in Atlanta...

that would be whether...

...he had a lot fewer minutes and touches when he played there.

as most 1st and 2nd year players do...

Now he is a significant player on a division champion team after being a bit player on a lousy team.

hmmm... let's see... so diaw's a significant player now but was just a bit player then? this 05-06 season diaw is playing the 3rd most minutes on a team with a roster of 11 players having played at least 100 minutes each. in 03-04 on the hawks (as a rookie yet) diaw played the 4th most minutes on a team with a roster of 17 players having played at least 100 total minutes each. that means that 13 other players on that 03-04 hawks team played less total minutes than did diaw, and of those 13 a full 10 played 400 less total minutes than did diaw...

must be an awful lot of "bit" players in the nba, huh?...

Would you ascribe more significance to the touches of a 35 min. a game player who likely is playing a lot against the other team's starters versus the touches of a 25 min a game player...

a touch is a touch - there are players that play similar minutes but one may get 4-5 more total touches than the other. does that make the touches of the player with the fewer touches any less important? of course not - it's what a player does with his touches that is important...

...who likely played a lot against reserves?

again diaw played the 4th most total minutes on the 03-04 hawks, and started 37 games. i'm willing to go out on a limb here and say he played against opposing starting players more often that season than 3/4 of the rest of the players on that team...

for that fact, can you show me even one team - just one - in nba history where the player playing the 4th most total minutes on that team played "alot" against reserves and not mostly against the opposing teams' starters? or does your definition of "alot" mean just a small part of a player's total minutes played?...

For whatever reason, Atlanta either didn't see much value in what he was doing or thought that the players ahead of him in the rotation contributed more value (perhaps such evaluations are why they miss the playoffs year after year).

if the hawks didn't see much value in him, why as a rookie did they play him 1919 total minutes? fyi the percentage of all nba rookies to play as much as 1919 minutes in their initial campaign in the league is small, so on the contrary the hawks must have seen something they liked in him to play him that much. in actual fact diaw played more total minutes in 03-04 than over half the rookies chosen before him in the 2003 draft....

sounds to me like atlanta liked him plenty - more than a number of other teams liked their rookies that season, because only 7 more players in that entire 2003 draft played more total minutes in the league that 03-04 season than did he (yet he was the 21st pick that year)...

they simply gave him up in a trade for a player (joe johnson) who was just a year older but had just shot better from 3pt range than any player had in the league in a decade...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mike G



Joined: 14 Jan 2005
Posts: 3623
Location: Hendersonville, NC

PostPosted: Mon Apr 17, 2006 6:30 am Post subject: Reply with quote
For whatever reason, Bob is obfuscating some facts to make some point (whatever that is.)

In '04, Diaw was indeed 4th in minutes for the Hawks. Thanks to lots of player movement, guys like Abdur-Rahim, Ratliff, Glover, Sura, Collier, and Przybilla all played more minutes per game -- while with the Hawks -- than did Diaw. So at any given time, he was about 6th in the rotation, at 25 mpg.

In any event, Diaw did not go from 2003-04 to 2005-06, any more than you or I did. In 2004-05, which is last season (the year immediately preceding this season), Diaw's minutes dropped to 18 per game. His production rates were down, and on an even-worse team, he dropped to 10th in MPG. This is a "bit player on a lousy team".

From there he went to Phoenix. He did not stop by 2004 and pick up those numbers.

Still, in his last season in Atlanta, he started 25 of his 66 games. So he played more 'vs starters' minutes than most 18-mpg players. My formula for such things estimates that last year, 58% of his opponents were starters; this year, I estimate 73%.

Atlanta did 'see something' in Diaw, starting him in 38% of his games. But they didn't seem to get it out of him. His numbers, to me, suggest a 'replacement level' player; hardly the triple-double threat he now is.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
bchaikin



Joined: 27 Jan 2005
Posts: 690
Location: cleveland, ohio

PostPosted: Mon Apr 17, 2006 1:51 pm Post subject: Reply with quote
For whatever reason, Bob is obfuscating some facts to make some point...

how is he obfisc... obfuska...obfesca...oblitera...ggrrr... making any facts obscure? perhaps you should re-read the thread, this time more closely....

...(whatever that is.)

oooo...must be some alterior motive to debating the facts presented in this discussion, eh?...

the following statement was made:

Now he is a significant player on a division champion team after being a bit player on a lousy team. Would you ascribe more significance to the touches of a 35 min. a game player who likely is playing a lot against the other team's starters versus the touches of a 25 min. a game player who likely played a lot against reserves?

diaw did not play 25 min/g in 04-05, as you said just 18 min/g. he did play 25 min/g in 03-04 his rookie season, so this statement is refering to his rookie season, and intimating he was a bit player...

so tell me again, what presented facts have i obfuzkated?...

In '04, Diaw was indeed 4th in minutes for the Hawks. Thanks to lots of player movement, guys like Abdur-Rahim, Ratliff, Glover, Sura, Collier, and Przybilla all played more minutes per game -- while with the Hawks -- than did Diaw. So at any given time, he was about 6th in the rotation, at 25 mpg.

combined sura, collier, and przybilla played less minutes than did diaw. combined they did have more starts than diaw (46 to 37), but from what i could find diaw started pretty much every game from early january to mid march for the hawks in 2004, and was then typically the reserve with the most minutes played after that for those games in which he did not start. he was also starting for the team before rahim and ratliff were traded, so it looks like the hawks gave up on dion glover and played diaw at SG (or SF depending on where stephen jackson played) alongside theo ratliff, shareef abdur-rahim, jason terry, and stephen jackson, then alongside jackson and terry prior to sura's arrival...

that doesn't sound to me like a player who was 4th on the team in total minutes playing "alot" of his minutes against reserves - starting for two straight months and then being the top reserve off the bench tells me he got plenty of seasoning against top players in the league. i find that far from being a "bit" player...

so instead of going from "..a bit player..." to someone who has "...entirely changed his game..." and made "...this incredible leap...", i think that by looking closely at the stats you'll find phoenix now has a player who despite switching from the backcourt to the frontcourt is now rebounding only slightly better than he did his rookie season, and is doing per touch (shoot with 1/5 of his touches, pass with 2/3, get fouled on 1/20) virtually the same thing in each of his 3 seasons, meaning he is and has been a passer first...

players playing substantial minutes (1000+ per year if not more) typically increase their touches/min early in their careers (2nd to 3rd and 3rd to 4th seasons), improve their shooting, and reduce their turnovers. to his credit what diaw has done has been to his increase his touches in consecutive seasons (his 1st to 2nd and his 2nd to 3rd) much more than is typical, and reduce his turnovers much more than is typical, while also improving his D...

In any event, Diaw did not go from 2003-04 to 2005-06, any more than you or I did.

correct - and he did not play 25 min/g in 04-05 either, or are you simply trying to obfuscate the 35 min/g vs. 25 min/g statement presented?...

...in his last season in Atlanta, he started 25 of his 66 games. So he played more 'vs starters' minutes than most 18-mpg players...

i'd 2nd that motion...

Atlanta did 'see something' in Diaw, starting him in 38% of his games. But they didn't seem to get it out of him. His numbers, to me, suggest a 'replacement level' player; hardly the triple-double threat he now is.

he was just 21 years of age his rookie season, played the 4th most minutes on his team, and started 37 games. they certainly saw something - that's a heckuva lot more playing time than most 21st overall picks get in the nba, at least in the past couple of decades. not many 21 year olds light up the league...

at 22 he then got buried on a poor hawks team behind stalwarts antoine walker and al harrington who were both playing 40 min/g before walker was asked to relocate. at that young age he shot better overall than walker did, had the same rate of turnovers per touch, same rate of steals, and blocks, yet sat more than he did, and saw more minutes go to rookies josh childress and josh smith, who were drafted in higher slots (6th and 17th) than he was. must have been frustrating...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Carlos



Joined: 21 Jan 2005
Posts: 64
Location: Montevideo, Uruguay

PostPosted: Mon Apr 17, 2006 4:23 pm Post subject: Reply with quote
A few of Diaw more detailed numbers in 82games caught my attention, such as; his gradual improvement as a passer (assist per 48 min. went from 4.6 to 6 to 8.4 and assists/passing turnovers went from 2.5 to 2.6 to 5). The relative consistency of his rebounding (from 18.2 to 15.1 to 19.4). The more agressive offense from 03/04 to 04/05 (went from 4.1 FGA in 25 min. to 4.5 FGA in 18 min.). The big changes in his "shooting profile" ( went from shooting 48% from the inside to only 34% in his second year to 51% this year). Most of the numbers suggest that he was simply developing (as a young player should) but that something made him shoot much more from the outside that was advisable in 04/05 (probably Harrington presence at the SF spot), which in turn reduced his productivity and his minutes.

uthor Message
Doc319
Guest






PostPosted: Tue Apr 18, 2006 5:00 am Post subject: Reply with quote
The whole issue of giving a player responsibilities versus signing him on the basis of his skills is basically semantics and not central to what I am interested in here--the question of whether Diaw has improved or is simply doing the same things he always did but with more on court time. I would in fact say that a team would sign Shaq for the purpose of giving him scoring/rebounding responsibilities. Is it not Shaq's responsibility to score and rebound? A team wouldn't sign him with the idea that he would lead the squad in steals or assists. For a historic example of a player who had varying responsibilities on different teams, Wilt Chamberlain scored 50 ppg one season and later in his career took on more playmaking responsibilities and drastically curtailed his FGAs. Just to prove to critics that he could still score if he wanted to (or had the same "game," to use the phrase currently in vogue), Chamberlain would break out for a 40, 50 or 60 point game once or twice a year. I don't think that Phoenix signed Diaw with the idea that he would be second on the team in assists. He was granted the opportunity to do that much ball handling and distributing when he showed that he was able to handle those responsibilities. Of course, Stoudemire's injury provided an opportunity for someone to play more minutes and Diaw has been the beneficiary of that opportunity. If Diaw were turning the ball over or simply unable to make passes that lead to assists than someone else would be given the responsibility of being the team's second ball distributor after Nash.

My post referring to Diaw's minutes appears to have caused some confusion. First I mentioned 25 minutes per game and then later in the post I referenced 25 minutes per game or less. So I see how different people could read that and interpret it in different ways. Bob thought that I was only referring to Diaw's rookie season, while Mike G looked at Diaw's body of work in Atlanta. What I was getting at is that in his two seasons in Atlanta he played 25 minutes per game or less (i.e., the phrasing I used later in that post), which is to say half of the game or less. Starters tend to play 32 minutes or more. Perhaps "bit player" was a bit harsh but Diaw was playing half the game or less for a bad team. Now he is playing 35 minutes a game for a division winner. No matter how anyone slices and dices the comings and goings of the flotsam and jetsam on Atlanta's roster, this is a significant upgrade in Diaw's oncourt time and it is hard to make a case that with that increased time he is playing less often against starters and starting quality players.

Diaw's minutes went down from his first year to his second year. Perhaps this was because of injuries, the shifting of personnel on the team or other reasons, but the decrease in minutes does not fit in with the theory that his game was developing as a young player's game would be expected to naturally develop. The "developing" has happened this year in Phoenix--significant increase in minutes and a corresponding increase in production. Most people seem to believe that Diaw has improved as a player. Bob's contention is that he was doing the same (or similar) things all along--on a per touch basis--in Atlanta. My question is, whether we look at his 25 mpg season or his 18 mpg season or the body of work of both seasons, does the sim take into account two possible factors: one, it is easier to maintain a given rate of production in a smaller amount of minutes (less fatigue, no foul trouble to possibly curb aggressiveness, etc.); two, a player who plays lesser minutes is likely on the court with a lower caliber of opposition players, making it easier to pass, shoot, rebound, etc. For the first point, consider another Sun, Eddie House, as an example. He scores roughly 10 ppg in roughly 17 mpg, providing "instant offense" off the bench, usually against the other team's second unit. Would the sim assume/project that if he played 34 mpg that he would average 20 ppg? Getting back to Diaw, prior to this season would it have been reasonable to assume that he could maintain his per minute rates in Atlanta with significantly increased playing time? Speaking of Diaw's rates, according to my calculations in his two Atlanta seasons he scored .21 points per minute, averaged .16 rebounds per minute and .106 assists per minute. This year he has nearly doubled his per minute scoring (.38), significantly increased his per minute assists (.17) and improved his per minute rebounding as well (.19); these numbers don't include the last couple games, so they might be off slightly, but it seems to me that Diaw is not only playing more minutes on a better team but he is scoring and passing at a much higher rate. By what calculation does the sim determine that he is doing the same things that he has always done on a per touch basis?

Again, the bottom line question that I am trying to answer is if Diaw has improved or is in fact as Bob suggests simply doing roughly the same things that he did per touch in Atlanta. My followup question to that is, if in fact Diaw basically has the same "game" that he had in Atlanta, how many other 25 mpg or less players are there in the NBA who could be transplanted to a good team and play 35 mpg while being second on the team in assists? I'd be interested to see a list of candidates, so that we can follow their careers and see how common or rare this in fact is. Anyone who could predict this kind of thing with a high degree of accuracy would be invaluable to a team; how many GMs would love to have a list of guys who are languishing on bad teams playing 25 or less mpg who could in fact play 35 mpg on a division winner without any loss of per minute production?

--David Friedman
http://20secondtimeout.blogspot.com/
Back to top

Carlos



Joined: 21 Jan 2005
Posts: 64
Location: Montevideo, Uruguay

PostPosted: Tue Apr 18, 2006 7:46 am Post subject: Reply with quote
My take is that Diaw was really improving in Atlanta during his second year, but that Harrington's arrival forced the Hawks to play him more at SG or PG, positions where his lack of shooting range hurted him a lot, and that "masked" his improvement. He had already shown signs of improvement before arriving to Phoenix, too, playing better in summer league and doubling his production with France's national team (voted as team's best player, over Parker and Pietrus).
Back to top
View user's profile Send private message
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Tue Apr 18, 2006 5:11 pm Post subject: Reply with quote
"how many other 25 mpg or less players are there in the NBA who could be transplanted to a good team and play 35 mpg while being second on the team in assists? I'd be interested to see a list of candidates"

Here are some names that came to mind as possible fits with these criteria: Beno Udrih, Jasikevicius, maybe Milt Palacio, John Salmons and Jarett Jack. Combo guards all but I could see bigger things including assists per game possible for most of them if in a different situation.

Already mentioned Luke Walton is the main big and then Deng and possibly D Granger if playing in a Paul Pierce way. (Joey and Steven Graham have some size/strength enough for SF or maybe a little PF coverage. Joey more offensive talent, Stephen more willing to pass, perhaps to mask not being that hot a shot right now.)

Further down the line but one case perhaps worth a flyer footnote might be lightly used so far Donta Smith. Diaw size body but more athletic? with some passing abilty shown in limited NBA time and in D league.

If you looked for a Diaw mix of skills in the draft, after Bargnani, Thabo Sefolosha sounds somewhat interesting in this vein http://draftexpress.com/viewprofile.php?p=280
6-7 205, guards best player on court, including (European) big men, rebounds, with handles,intelligent, hardworking etc. Sound like he could play SF especially after some weight work here.

Jerry Stackhouse is a pretty good passer from the SG and SF and has played an occasional minute at PF. Now 32 if quickness became a defensive issue he might play more forward and more inside like the last hoorah Jordan. If traded to a different team and they needed it he might still be able to play 35 minutes and be a #2 passer.)

(Rick Fox skills/usage had some resemblance to Diaw, playing PF at times when needed. If you looked at a long view of history there are more 6-6 to 6-8 players who played inside and made contributions in many ways, some at much higher levels such as Barkley and Baylor)

Last edited by Mark on Wed Apr 19, 2006 3:37 pm; edited 1 time in total
Back to top
View user's profile Send private message
Doc319
Guest






PostPosted: Tue Apr 18, 2006 11:43 pm Post subject: Reply with quote
That's a pretty interesting list, Mark. The only guy I would disqualify is Stackhouse; he has already been a 35 mpg performer, so the only question with him would be is he still healthy enough to be a productive 35 mpg player. Diaw never was a 35 mpg player until this season and I believe that the same is true of the other players that you listed. What particular statistical tendencies/similarities among these players or with Diaw led you to select them? I understand Bob's original post about Diaw to mean that Diaw's performance this year is not unexpected and would in fact have been predictable, assuming that the predictor knew how much Diaw's minutes would increase. I still don't understand how his "per-touch" numbers are supposedly the same or similar when his per-minute scoring, assists and even rebounds have all increased. It seems to me that it would have been a bit of a reach before this season to predict that Diaw would be this productive, which is why I am curious if such a "prediction" can be duplicated--i.e., is it possible to use the same techniques that "would have" predicted Diaw's productivity with the Suns to identify the next "Diaw," as opposed to waiting until he shows up and then saying that his arrival was easy to predict.

--David Friedman
http://20secondtimeout.blogspot.com/
Back to top

bchaikin



Joined: 27 Jan 2005
Posts: 688
Location: cleveland, ohio

PostPosted: Wed Apr 19, 2006 12:45 am Post subject: Reply with quote
the bottom line question that I am trying to answer is if Diaw has improved or is in fact as Bob suggests simply doing roughly the same things that he did per touch in Atlanta.

are you under the impression this is an either/or scenario? these are not mutually exclusive events - players increase (and decrease) their touches/min all the time yet maintain what they do per touch...

I still don't understand how his "per-touch" numbers are supposedly the same or similar when his per-minute scoring, assists and even rebounds have all increased.

break out your abacus - if a player gets 32 touches in a game playing 36 minutes and takes 8 shots, that's 1 shot for every 4 touches and 0.89 touches/min. if the next game he plays 36 minutes and gets 36 touches and takes 9 shots, that's 1.00 touches/min but still 1 shot for every 4 touches. take this same concept and extend it to consecutive seasons as opposed to just consecutive games...

It seems to me that it would have been a bit of a reach before this season to predict that Diaw would be this productive, which is why I am curious if such a "prediction" can be duplicated--i.e., is it possible to use the same techniques that "would have" predicted Diaw's productivity with the Suns to identify the next "Diaw," as opposed to waiting until he shows up and then saying that his arrival was easy to predict.

look at the historical stats - there are plenty of examples of this...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 19, 2006 2:19 pm Post subject: Reply with quote
Donta Smith 2006:

TS% 58%, assist rating 21, rebound rating 6, utilization rating 13


Donta Smith 2005:

TS% 48%, assist rating 20, rebound rating 7, utilization rating 16

His minutes were cut in half. Shooting % improvement and A/TO improvement from 1.5 to 2.3 but some treading water otherwise (like the Diaw did that). Does he also have more potential? I dont claim to know, just thought he was a useful case for the topic. He has had a good last 10 games 70% FG, 50% higher scoring rate per 48 and tendex. Will he get to and show it in Atlanta or will he have to go elsewhere like Diaw did?


Boris Diaw 2006:

TS% 56%, assist rating 31, rebound rating 11, utilization rating 17


Boris Diaw 2005:

TS% 48%, assist rating 26, rebound rating 8, utilization rating 15


I wondered if Atlanta felt Diaw was expendable not only because Joe Johnson was coming but they had Childress and both of the Smiths.
I knew Atlanta loved 6-6 to 6-8 guys from previous study, so if you were looking for future Diaws they seemed a good place to look.

If Bob has Donta Smith's per touch analysis handy or easily produced, I would appreciate the assist. Looking at these measures I dont think it would be too surprising if Donta Smith given the time and therole could equal or exceed 2005 Diaw next year or the year after and might reach 2006 Diaw levels at some point. Per 48 minutes Donta is 13 pts 5 rebs 4 assists rounded off, while last year Diaw was 13 pts 7 rebs 6 assists per 48. Anyone who has watched him enough to offer a more informed opinion that would be welcome.

His draftnet bio said in part :

"Strengths: Wonderfully built pro ready body standing at 6-7 and weighing 230 pounds … Because of his strength he is a great finisher after contact occurs in the lane … Ball handling ability is very refined for a player his size … Add in his quick first step and he is a threat to take defenders off the dribble … Extremely athletic leaper capable of impressive forays to the basket … Has tremendously quick hands that enable him to disrupt passing lanes … Great rebounder for his size who understands how to use his body and athleticism under the glass ..."


P.S. I knew Stackhouse violated the under 25 minute criteria for career but he has been close to that level for awhile and thinking of Diaw a wing going more inside, I just thought is there any real reason Stackhouse couldnt do it too if asked? I dont expect it, it might not be anymore value from him than playing normally but I think he could do it.

Will the Diaw example encourage more use of strong "wings" playing an inside postion? Starting officially at 4 or 5 may still be too much for most but being a 3/4 inside/outside seems modern, workable and we may see more. Especially is athletic PFs more often play the center position and the crop of shooting powerforwards (with 3pt skill) expands in coming decade as I assume it will from Europe but also from college as this is a viable path to an NBA job now. In this case the position designation isnt as important as the role.

If you have a modern center playing more offense (Amare) and a modern 4 shooting more outside (Marion) (space between the official bigs to avoid double team post shutdowns?) , then a 3 (Diaw) who can contribute to rebounding, inside defense, inside scoring (by slicing or posting a smaller man)and playmaking help seems to make sense fulfilling the roles of the past across the 3 positions in a new mix? Diaw more at SF next year? as starter or backup 3 along with some backup 4 while Thomas gets the backup 5 minutes? Will be interesting to see how D'Antoni choreographs Thomas/Amare S Marion/Diaw time and position and role next year.

Atlanta has Zaza I see as really a PF turned C (in part due to the tragedy with Collier), Marvin Williams as the shooter (it is hoped he will further develop his shooting skills in combo with some inside game) or SF playing PF, and lots of guys who can contribute on inside work from the 3 spot (and big 2 or backup 4).

Last edited by Mark on Wed Apr 19, 2006 5:41 pm; edited 2 times in total
Back to top
View user's profile Send private message
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 19, 2006 4:51 pm Post subject: Reply with quote
I thought about trying manual calculations of touches for current year but after reviewing the formula, just using the database is the way to go.

Last edited by Mark on Wed Apr 19, 2006 5:35 pm; edited 2 times in total
Back to top
View user's profile Send private message
Eli W



Joined: 01 Feb 2005
Posts: 402


PostPosted: Wed Apr 19, 2006 5:03 pm Post subject: Reply with quote
If you download Bob's database from the APBR site it is updated with 04-05 stats. It's the link called Statistical Database under the big APBR logo at the top here:

http://hometown.aol.com/bradleyrd/apbr.html
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 19, 2006 5:12 pm Post subject: Reply with quote
Thanks for the tip. I had an old 2003 copy of the database in my files. The bballsports.com link says thru 2004 but I wasnt sure if that mean thru 2003-4 or 2004-5, I initially thought it meant the former. But using the link you mentioned I will look at the thru last season db and look forward to the next update thru this year.


I see Donta Smith last year at 0.94 touches, 28% shoot, 56% pass 6% TO, 10% Fouled.

Add: and this year I now see Donta Smith 0.69 touches, 29% shoot, 59% pass 2% TO, 10% Fouled. Touches down by more than 25% but much fewer turnovers and more passes. Seems like player improvement and role reduction. Perhaps better is ahead.

Last edited by Mark on Wed Apr 19, 2006 8:19 pm; edited 1 time in total
Back to top
View user's profile Send private message
bchaikin



Joined: 27 Jan 2005
Posts: 688
Location: cleveland, ohio

PostPosted: Wed Apr 19, 2006 6:12 pm Post subject: Reply with quote
If you download Bob's database from the APBR site it is updated with 04-05 stats. It's the link called Statistical Database under the big APBR logo at the top here:

http://hometown.aol.com/bradleyrd/apbr.html

i haven't updated the APBR stats database yet with 05-06 numbers, but the superDB at www.bballsports.com has recent nba stats...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 19, 2006 6:40 pm Post subject: Reply with quote
Ok. I will download that one too.

(Donta Smith year to year touches based data comparison shown above in previous post.)

Last edited by Mark on Wed Apr 19, 2006 8:23 pm; edited 1 time in total
Back to top
View user's profile Send private message
Eli W



Joined: 01 Feb 2005
Posts: 402


PostPosted: Wed Apr 19, 2006 6:41 pm Post subject: Reply with quote
Thanks for the tip Bob. I didn't realize you had a version of the database available with updated stats for the current season.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 19, 2006 8:47 pm Post subject: Reply with quote
Walton this year 1.04 touches per minute, 23% shot 5% fouled 6% TO 66% pass
Walton last year 1.07 touches per minute, 22% shot 5% fouled 7% TO 66% pass

Diaw this year 1.48 touches per minute, 20% shot 5% fouled 4% TO 71% pass
Diaw last year 1.04 touches per minute, 22% shot 5% fouled 7% TO 66% pass

Walton last year and this year nearly identical to Diaw last year.

(Stackhouse this year 1.26 touches per minute, 31% shot 10% fouled 6% TO 53% pass. Different, but maybe not as big as his earlier career.)
Back to top
View user's profile Send private message
Doc319
Guest






PostPosted: Thu Apr 20, 2006 4:38 am Post subject: Reply with quote
Bob, if I understand you correctly you are saying that Diaw's minutes and touches have increased this year but that what he does per touch in terms of shooting, passing, etc. has remained pretty much the same (other than the decrease in TOs and increase in blocks that you mentioned). So he has improved in the sense of being productive in a larger amount of minutes and is doing the same (or similar) things in terms of your per touch calculations.

--David Friedman
http://20secondtimeout.blogspot.com/
Back to top

dlirag



Joined: 30 Dec 2004
Posts: 29


PostPosted: Fri Apr 21, 2006 1:58 am Post subject: Reply with quote
Cutting TO from 7% to 4% looks like a major improvement.

Author Message
Mark



Joined: 20 Aug 2005
Posts: 807


PostPosted: Wed Apr 26, 2006 8:32 pm Post subject: Reply with quote
HoopsHype.com Interviews Boris Diaw:

"When they brought you to Phoenix, the coaching staff had ideas of playing you at the 1 and 2.

BD: It was the 1 actually…

Then you ended up playing most games at the 4 and 5. Were you surprised with this opportunity? Were you aware before of your ability to play those positions?

BD: I don’t know. I am always ready to play anything you ask me to play (laughs). That’s what I have always been doing. Just ask me, and I will try to do it.

So you played at the 4 and 5 in your career before this season?

BD: Not really. Only in spurts, when that was needed by my teams.

How do you explain your ability to successfully play those positions?

BD: I think it is a bit because of the versatility and the fact that I try to do different things on the court regardless of the position in which I am playing.


http://www.hoopshype.com/interviews/diaw_mustafic.htm

Diaw at the one... his brief experiment at that spot in Atlanta showed no signs of being his right spot. His by position chart showed him getting better in most ways as he moved higher on the position ladder there up thru SF. His few minutes at PF werent enough to really analyze.

Could he use the confidence he gained this year and play some time at 1 or 2 next year in addition to time at 3/4/5?

I assume the Suns would like to get 90+ minutes out of Stoudamire, Marion, K Thomas. If T Thomas stays for 20 (he may not due to rebounding price) and J Jones for 20 (not certain or essential), I guess Diaw could still 24 minutes inside. Behind Nash and Bell, House and Bell are decent but not amazing. Maybe Diaw gets 12 more minutes on the perimeter on some nights, depending on how the little backups are doing/matchups?

It would be interesting if Diaw could play a big PG. He did it in France.
His FG% on jumpers went up from 34% last year to 41% this year. Still a liability on the perimeter but if Jones and Marion tend to drift outside there might still be room for him inside even from a perimeter spot.


P.S. I found some Donta Smith compared to Diaw talk here http://www.hawksquawk.net/forums/showfl ... mber=70911
7-1 wingspan.

nbdl experience:

"Originally assigned to the RimRockers on December 2, 2005, Smith played in 29 games (22 starts) with Arkansas, averaging 13.9 ppg, 4.3 rpg, 3.0 apg and 1.7 spg in 33.2 mpg (.568 FG%, .323 3FG%, .731 FT%). He ranked 5th in the league in field goal percentage, 6th in steals, 18th in assists, 20th in scoring and tied for 32nd in rebounding."

(Diaw this year in NBA- 13.3 pts, 6.9 rebs 6.3 assists 53% FG 73%Ft, 0.7 steals)

this earlier comparison is interesting/funny in retrospect
http://www.hawksquawk.net/forums/showfl ... Post109540

On Donta Smith: "When he signed Smith, Pitino likened him to Indiana Pacers standout Ron Artest.

"We thought we were getting a kid who could be a major defensive stopper and a great rebounder with great athleticism," Pitino said.

I hadnt thought about Artest in comparison to DSonta Smith or Diaw but it works in many ways.