The Spurs' varied skills on defense..(or lack thereof) ..from 48MoH

[Spursfanfromafar]

The problem with Adjusted +/- is twofold. Because it relies on a myriad of other "non-boxscore" stats and regressions, it's much more prone to noise due to small sample sizes. I recall when it was introduced in 82games.com, they even pointed out that a full regular season was likely not a good enough sample size. The second problem is pretty much the same as regular +/-. There's an inherent skewing when player are paired (one very productive and one not so productive) and largely share the same lineups.

Now, don't get me wrong. I think Matt has shown his value in the regular season. Just pointing out what's the deal with Adjusted +/-, and why it's also an almost unusable stat to gauge playoff performances.

1) You are right about sample sizes. I guess, the better APM sites use atleast 2 seasons of data - such as what is shown up on basketballvalue.com

2) The second point I disagree with - the purposes of having adjusted APM over situational PMs as described by Dan Rosenbaum are to whittle away the effect of pairing with productive/non-productive players. The regressions are so done to account for various pairings/lineups by comparing with that for a "replacement player". Don't know if I am saying that accurately, but that is the general idea.

[ElNono]

2) The second point I disagree with - the purposes of having adjusted APM over situational PMs as described by Dan Rosenbaum are to whittle away the effect of pairing with productive/non-productive players. The regressions are so done to account for various pairings/lineups by comparing with that for a "replacement player". Don't know if I am saying that accurately, but that is the general idea.

Additionally, the estimates suffer from the issue of skewed sampling – the fact that most players usually find themselves on the court in the company of certain teammates and not others. As a result, it can be difficult to accurately tease out the individual effects of two players who almost always appear on the court together. Rosenbaum and others have outlined different ways of addressing these issues, most notably using multiple years’ worth of data and augmenting regression results with additional analyses based on box score statistics.

Link

From 2007, when 82games.com started to do APM mid-season. Includes link to the original 2004 Rosenbaum article you posted.

[therealtruth]

Because Bonner scores so well in those, there is something about him being good in the regular season as the opposition is not paying enough attention to his strengths. Unfortunately that luxury is not available in the playoffs, where his weaknesses are better taken advantage of and hence the drop in his value and production.

Against good defensive teams Bonner is a liability. They'll negate any floor spacing effect he has. Then you have to ask yourself if you're just playing Bonner for his defense.

[Spursfanfromafar]

Additionally, the estimates suffer from the issue of skewed sampling – the fact that most players usually find themselves on the court in the company of certain teammates and not others. As a result, it can be difficult to accurately tease out the individual effects of two players who almost always appear on the court together. Rosenbaum and others have outlined different ways of addressing these issues, most notably using multiple years’ worth of data and augmenting regression results with additional analyses based on box score statistics.

Link

From 2007, when 82games.com started to do APM mid-season. Includes link to the original 2004 Rosenbaum article you posted.

Yes. In the 2004 article itself, Rosenbaum adjusts his Pure APMs with OLS estimates based on box score stats, fact that Barzilai (author of basketballvalue.com) himself mentions in the last line you quote him...which is the key for me -

"Rosenbaum and others have outlined different ways of addressing these issues, most notably using multiple years’ worth of data and augmenting regression results with additional analyses based on box score statistics. "

I think adjusting for box score based stats and APM - i.e. taking a middle path between PER/win Shares and APM, one can say that Bonner is still very productive as compared to other bigs in the league.

But we are quibbling. As much as the stats say so good about Bonner, they mean little in the games that matter the most - the playoffs. That much, we do agree.

[OrEmuN]

Just thought of a theory abt Bonner after looking at the article.
I wonder whether because Bonner is such a terrible defender by visual sight that everyone who are going up against him in regular season tend to underestimate him and throw up any shots. That results in bad shot selection and hence making Bonner's stats look better than his actual defense.
In addition, when the playoffs come, everyone are more focused and would not make the same underestimation.

[Darkwaters]

On the other end of the spectrum, DeJuan Blair is simply a terrible defensive player. I love DeJuan’s grit and hustle, but in 2011, DeJuan was the worst defensive starter at the center position in the entire league by DPPP. In 2012 he isn’t the worst, but he’s close. Despite defending on 4 possessions fewer than Tim Duncan, he’s allowed 56 more points scored directly on him over the course of the year. That’s quite a gap. And while part of that is Tim’s overall stellar defense, a lot of the blame has to be put on DeJuan himself — while he’s decent on the P&R, he’s poor in isolation and a travesty at recovering on spot-up shooters. Which, by the way, is why he plays center so often. I’ve been asked by friends who don’t follow the Spurs why DeJuan (a 6’7″ player) is forced to play center so often. It isn’t necessarily because Pop wants to do it, it’s because if he cross-matches onto a power forward he gives them a wide open look at a spot-up jumper virtually every time. He’s no great shakes in the post, either, but at least he’s slightly above league average. When you put the whole package together, though, you have a player who’s league average at best in most defensive categories, and who unfortunately averages out as a big liability on the defensive end. Which is a shame, because again — I really love DeJuan, and I really wish he was just a few inches taller and a few pounds lighter. But as of this point, his career has been awful on the defensive end, and it’s really aggravating to watch him work.


Read more: http://www.48minutesof .com/boris...#ixzz1qIj6wgYh

Validating what most of us have been saying all along.

[GSH]

Apparently he has already done that.



What the writer says is that, Bonner has some good skills and some bad ones and the bad ones stand out.

To those who question +/- stats, please note that adjusted +/- stats are very useful as they do not rely on box scores or some such, but actual regression analysis comparing production over the average "replacement player". Adjusted +/- are very effective, as compared to box score based stats therefore.

Because Bonner scores so well in those, there is something about him being good in the regular season as the opposition is not paying enough attention to his strengths. Unfortunately that luxury is not available in the playoffs, where his weaknesses are better taken advantage of and hence the drop in his value and production.

With respect, I appreciate what you say about adjusted +/-. The goal and the idea behind it are both admirable. The reality of the stat - not so much. I don't want to go into a page-long discussion, but I will say this much: any statistical model that determines that two of the top 20 players for the year were Brian Cardinal and Shawn Bradley is pretty close to worthless.

The idea was to create a model that would contain most of the obvious names, such as Duncan, Kobe, LeBron, etc. - but also contain a few "sleepers". In other words, players that are of more value to a team than their stats would indicate. Without that, what good would the model be? The model was supposed to allow teams (like Cleveland) to do a "Money Ball" act, by allowing them to identify the great contributors that could be had cheaply and easily. It didn't work.

The statisticians use an old psychic trick when evaluating the model. When a player has a high adjusted +/- and plays very well, they say it was a successful prediction. When a player has a high adjusted +/- and plays like crap, they call it "noise" in the data. In other words, it's a great stat for looking backwards. It's not worth much for predicting. And, as I said, most of the names on the list are the players we all know are difference makers.

In 03-04, Brian Cardinal was supposedly the 14th biggest difference maker in the league, tied with Paul Pierce. That's all you really need to know about adjusted +/-.

[TheSkeptic]

Validating what most of us have been saying all along.

Is that another endorsement for bringing Blair off the bench I hear?

[Spursfanfromafar]

With respect, I appreciate what you say about adjusted +/-. The goal and the idea behind it are both admirable. The reality of the stat - not so much. I don't want to go into a page-long discussion, but I will say this much: any statistical model that determines that two of the top 20 players for the year were Brian Cardinal and Shawn Bradley is pretty close to worthless.

The idea was to create a model that would contain most of the obvious names, such as Duncan, Kobe, LeBron, etc. - but also contain a few "sleepers". In other words, players that are of more value to a team than their stats would indicate. Without that, what good would the model be? The model was supposed to allow teams (like Cleveland) to do a "Money Ball" act, by allowing them to identify the great contributors that could be had cheaply and easily. It didn't work.

The statisticians use an old psychic trick when evaluating the model. When a player has a high adjusted +/- and plays very well, they say it was a successful prediction. When a player has a high adjusted +/- and plays like crap, they call it "noise" in the data. In other words, it's a great stat for looking backwards. It's not worth much for predicting. And, as I said, most of the names on the list are the players we all know are difference makers.

In 03-04, Brian Cardinal was supposedly the 14th biggest difference maker in the league, tied with Paul Pierce. That's all you really need to know about adjusted +/-.

Actually, Cardinal indeed was so productive in his contract year in 03-04 that the next year he got a major contract and he never lived upto that production again.

And basically I disagree with everything you say - "psychic trick" / "looking backwards"/ "not worth much for predicting" etc.

Given tools in our disposal - eye test, stats based on box score numbers, stats based on APM and stats based on overall APM and so on.. the latter are the best available. These are now being supplemented by what Synergysports does - capturing every possession and tabulating them under categories. This makes it much much more effective to predict, compare and tabulate than just mere scouting opinion.

It is no rocket science to know that the best teams (including the Spurs) today are employing good statisticians and number crunchers who can supplement their skills with basketball observation.

So, you are wrong about the efficacy of stats based observations such as APM.

[ElNono]

Yes. In the 2004 article itself, Rosenbaum adjusts his Pure APMs with OLS estimates based on box score stats, fact that Barzilai (author of basketballvalue.com) himself mentions in the last line you quote him...which is the key for me -

"Rosenbaum and others have outlined different ways of addressing these issues, most notably using multiple years’ worth of data and augmenting regression results with additional analyses based on box score statistics. "

I think adjusting for box score based stats and APM - i.e. taking a middle path between PER/win Shares and APM, one can say that Bonner is still very productive as compared to other bigs in the league.

But we are quibbling. As much as the stats say so good about Bonner, they mean little in the games that matter the most - the playoffs. That much, we do agree.

To me the kicker is that APM's supposed added-value (compared to unadjusted) is that it uses much more information that's not related to the boxscore. And the reason is fairly simple: unadjusted (boxscore) plus minus is a clearly a lineup-combination stat, not a player-centric stat. Trying to turn it into a player-centric stat requires much more info, but also a much larger sample size.

Yet, since it's basically still based off lineup-combinations (can't get around that, since that's how the game is played), it's still prone to skewing on player combos that overlap timelines, which can happen a lot, especially on teams that have established rotations.

The solution? Go back to boxscores to get hard numbers to isolate the individual production. It kind of diminishes the added-value, IMO.

[ElNono]

Just thought of a theory abt Bonner after looking at the article.
I wonder whether because Bonner is such a terrible defender by visual sight that everyone who are going up against him in regular season tend to underestimate him and throw up any shots. That results in bad shot selection and hence making Bonner's stats look better than his actual defense.
In addition, when the playoffs come, everyone are more focused and would not make the same underestimation.

To me it's pretty simple:

Regular season = getting YOUR team ready. You see different opponents every other day, playing different sets and players. There's no point to obsess over them other than read a scouting report, etc. This is where having experienced players pays off too.

Playoffs = lock in on your opponent. You're only going to be playing that team and you either are going to beat them or go home. You create your gameplan by studying the opponent's players and try to find the weak links where you can maximize your chances of winning. When one of your bigs is Tim Duncan and the other is Matt Bonner, it's a no-brainer to pick which one to attack with both bigs and penetrating guards.

[therealtruth]

To me it's pretty simple:

Regular season = getting YOUR team ready. You see different opponents every other day, playing different sets and players. There's no point to obsess over them other than read a scouting report, etc. This is where having experienced players pays off too.

Playoffs = lock in on your opponent. You're only going to be playing that team and you either are going to beat them or go home. You create your gameplan by studying the opponent's players and try to find the weak links where you can maximize your chances of winning. When one of your bigs is Tim Duncan and the other is Matt Bonner, it's a no-brainer to pick which one to attack with both bigs and penetrating guards.

It's not just that. In the playoffs you look over the statistical data and you see the Spurs are nearly unbeatable when Matt Bonner plays well and hits multiple 3's. You adjust your gameplan to take Matt Bonner out of the game. The Spurs run the same gameplan and then find the Matt Bonner is not that effective. Pop continues to trot him out 20 minutes a game despite this and the Spurs now have to win despite him.

[jjktkk]

It's not just that. In the playoffs you look over the statistical data and you see the Spurs are nearly unbeatable when Matt Bonner plays well and hits multiple 3's. You adjust your gameplan to take Matt Bonner out of the game. The Spurs run the same gameplan and then find the Matt Bonner is not that effective. Pop continues to trot him out 20 minutes a game despite this and the Spurs now have to win despite him.

Since Pop has more options now with Diaw, and Splitter, and Leonard and Jack, if he chooses small ball, he shouldn't have to overuse Bonner. You did notice the new additions while trolling didn't you?

[angelbelow]

Actually, Cardinal indeed was so productive in his contract year in 03-04 that the next year he got a major contract and he never lived upto that production again.

And basically I disagree with everything you say - "psychic trick" / "looking backwards"/ "not worth much for predicting" etc.

Given tools in our disposal - eye test, stats based on box score numbers, stats based on APM and stats based on overall APM and so on.. the latter are the best available. These are now being supplemented by what Synergysports does - capturing every possession and tabulating them under categories. This makes it much much more effective to predict, compare and tabulate than just mere scouting opinion.

It is no rocket science to know that the best teams (including the Spurs) today are employing good statisticians and number crunchers who can supplement their skills with basketball observation.

So, you are wrong about the efficacy of stats based observations such as APM.

APM alone is not enough to be effective enough to predict performance. Its one aspect that shouldn't be ignored, nor should it be interpreted as effect by itself. Like you mentioned, there are numerous ways to predict a players performance but I don't think APM is the most effective.

For example, against regular season compe ion Bonner does incredibly well. However, as someone else previous noted, Bonner's playoff APM is no where near his regular seasons. In addition, against teams like Chicago and Miami, Bonner's posts negative APM in the double digits. Therefore, I still think that APM is unpredictable and situational. But combined with context, can be effective in predicting the value of a players worth to a team.

[benefactor]

More of this? Bonner helps in the regular season. Bonner craps the bed in the playoffs. History and numbers back it up.

/thread

[Dr. John R. Brinkley]

If Bonner can't defend in the post in isolation then he can't defend in the post. Period. To me, that's the essence of post defense, and this aspect is crucial to success in the playoffs.

[diego]

those same stats say that Hickson is a better defender than Splitter.

Its not the same to pick up a rebound on a missed FT, than to grab a board in a packed lane, but statistically they are the same play. (speaking of which, the article describes Bonner's rebounding as "virtually non-existent", which is pretty damning; if you force a miss but cant grab the board, all that work was for naught. Rebounding is easily 50% of defense, if not more in the case of a "big", and Bonner is absolutely terrible at it).
Then you have the other problem: stats dont account for smart / stupid play or compe ive / cowardly play. So a player that doesnt take an open look and dumps it on someone else in worse position because he's scared to miss or leaves his teammate out to dry on D because he is stupid or lazy is statistically more efficient and has better numbers than the one that takes the shot and misses or the one that helped and got "scored on".
Then add the playoff factor and the psychological factors... how many times have we seen Bonner looking scared and lost in the PO, incapable of making a shot, getting a board, or contesting a shot?
Bonner isnt trash, but you'll need a lot more than some statistics to convince me he is a better defender/player than any of the others on the list.

[Richie]

Against good defensive teams Bonner is a liability. They'll negate any floor spacing effect he has. Then you have to ask yourself if you're just playing Bonner for his defense.

How does that work? Either you cover him on the 3 pointer and open up the lane for Tony/Manu, or you don't and leave yourself vulnerable for the kickout to him?

Obviously the better defenders will read it and be able to get out quicker to him to defend his shot if they get drawn in helping against a driving Parker or Manu, but I don't really see how a team 'negates floor spacing'.

[Lebowski Brickowski]

I've been through this mental gyration a few times before. Nobody can be as bad as Bonner is supposed to be, and still have positive +/- stats over a long period of time. It's a math function. If only bad things happen when the player is in the game, his +/- would be negative. Maybe his defensive ability has holes, but he obviously contributes more than enough to make up for that.

And then I remember the playoff series against Memphis.

I want the person who wrote this article to go back and look at those games, and THEN write a commentary on Bonner's post-up defense. I defended Bonner for a long time, but the images of that series are burned into my mind forever - and a lot of them involve Bonner. I can't say that he was worse than a pylon on defense, but I can't say that he was any better. If he performs in this year's playoffs, I'll give him credit. But not until then.

Tim looked almost as bad.

[GSH]

Actually, Cardinal indeed was so productive in his contract year in 03-04 that the next year he got a major contract and he never lived upto that production again.

And basically I disagree with everything you say - "psychic trick" / "looking backwards"/ "not worth much for predicting" etc.

You can't use it to decide which players are the best, and therefore the best value. Brian Cardinal wasn't. Shawn Bradley wasn't. And the others, like Duncan, Lebron, Kobe, etc. are so obvious, you don't need a statistical model to figure it out. If you can't use it to predict, it's only good for looking backwards.

Given tools in our disposal - eye test, stats based on box score numbers, stats based on APM and stats based on overall APM and so on.. the latter are the best available. These are now being supplemented by what Synergysports does - capturing every possession and tabulating them under categories. This makes it much much more effective to predict, compare and tabulate than just mere scouting opinion.

So there's nothing out there besides mere opinion and APM? And you say that stats based on APM are the best, and that makes it so? Wow, with logic like that, you can never be wrong.

It is no rocket science to know that the best teams (including the Spurs) today are employing good statisticians and number crunchers who can supplement their skills with basketball observation.

And only two or three, I believe, are even attempting to use adjusted +/- numbers. There's a reason for that.

So, you are wrong about the efficacy of stats based observations such as APM.

I like that - you give an opinion, and that makes me wrong. Works best when you just throw away all the facts.

First of all, I never said Cardinal was horrible, or that he didn't earn one good (for him) contract. But the model said that he was the 16th best player in the league, and that is just ridiculously stupid. And I don't think even you believe it. Which means the model is deeply flawed. (BTW - the 06-07 APM list showed Antonio McDyess as the 20th best player in the league. He put up 8.1 points, and 6.0 boards. Was he a good player? Sure. Was he anywhere near the 20th biggest contributor in the league? Not even McDyess would have said that.)

Be careful, telling stories like that, when you don't know anything about what happened. Cardinal got that contract because Heisley was ing at Jerry West about not getting players signed. West got pissed, and basically said "You want a player signed? I'll sign a player." He called Cardinal's agent and offered him the full MLE, and signed him that day - and everyone in the league was shocked that West gave Cardinal such a stupid contract until it got out what had happened.

I never said that I don't believe in advanced statistics, just not this one. Bottom line, the APM tells us that players like Duncan and LeBron are good. We knew that. And then there are the "surprise players" like Bradley and Cardinal, and the statisticians call that "noise". Most people don't see the predictive value in that. You can keep defending it if you want to, I guess. For the record, I've had this discussion with two of the math-heads who did this for real live basketball teams - and even they admitted to more problems with APM than you do. They didn't like it when I called it a psychic trick, either. But they didn't deny that they wouldn't hire one of the surprise players for millions of dollars, based on the idea that they are the biggest "contributors". And if you adjust out all the "noise", then it's just a list of the guys we knew were good already.

You're obviously a believer, so I don't expect to convert you, or even for you to listen to logic. But we discuss basketball here, and the things you say should at least be rebutted. Because they're wrong. The guys who created and maintain the stat don't even believe in it the way you do.

[Horse]

He's needed alot less now with all the depth we have and that's a perfect role for him come in and hit 3's that's it. I will say though his hustle has impressed me recently.

[Spursfanfromafar]

@ GSH,

The trouble with what you say above is that you don't really pinpoint problems with the methodology adapted in APM, but rather say something exterior to it to say that it is not good enough. That, IMO, is a poor way of saying that APM is not good enough.

So, when you say that all APM does is point out to us, who the best are and we already know who they are.. and that it throws up noise.. that is not quite a valid criticism. APM's job is not only to point out the top 20 or 30..but to also highlight comparisons across teams and for various players. So, if I want to get, say, a 12th man for my team and I need to check his stats, APM helps me far more than petty conventional stats. And it is in this light I found your statements about "psychic trick", "looking backwards", "not predictive" problematic.

I am not saying that APM does not have flaws; I am saying that it- supplemented with other info - is a more robust advanced statistics universe than most..

And I don't think if APM is used only by 2/3 NBA stats officials. The 3 that I know are/ (were recently) consultants for teams already - Barzilai, for e.g. consults for the Grizzlies; Steve Ilardi consulted (don't know if he does still) for the Suns; and the 82games guy also consults for some NBA team, IIRC. Wayne Winston used to consult for the Dallas Mavericks, IIRC. Rosenbaum for the Cavaliers...

Lastly, I know this is a bulletin board and that opinions are meant to be debated and rebutted. You needn't remind me about that.

[HarlemHeat37]

I had a Synergy subscription in 2010 and 2011, and unless they changed the method of rankings in 2012, then they still don't weigh the quality of opponent in the rankings..a player that guards opposing bench players and secondary bigs will probably have a superior ranking to a player that guards the top player on the opposing team, on a nightly basis..

For instance, in Malik Hairston's last season with the Spurs, Synergy had him ranked as the best isolation defender in the NBA..I was a Hairston fan, and he clearly had defensive potential, but he obviously wasn't the best 1 on 1 perimeter defender in the NBA..the numbers were skewed, as his opponents were generally weaker..

[dbestpro]

Bonner may be okay with post defense, but most of the offense that he sees is out on the floor as he tries to guard big perimeter players. He also makes as many physical and mental mistakes on rotation defense as any player on the team.

Most Spur fans are not so gullible to be sucked into one aspect of his defense and then declare him a defensive star.

[therealtruth]

How does that work? Either you cover him on the 3 pointer and open up the lane for Tony/Manu, or you don't and leave yourself vulnerable for the kickout to him?

Obviously the better defenders will read it and be able to get out quicker to him to defend his shot if they get drawn in helping against a driving Parker or Manu, but I don't really see how a team 'negates floor spacing'.

You explained exactly how it works. They defend Manu and Parker on the drive and prevent Bonner from getting open looks. The Grizzlies did a really good job of this.