SoTD: Bargnani's defensive numbers over 4 years

[supersub15]

best set of ballboy posts ever!

Boris has made an excellent statisitically based counter argument with a wealth of insight and analysis.

I respect Sub's findings as more than just anomolous based on the size of the data set. However I suggest that since Bargnani has been used more as a prime time player he has been ON vs other team's best and when he is OFF it's vs the other team's bench therefore when you see the ON/OFF numbers you must consider the nature of the data set as you are comparing 19-25 PER stars vs 10-15 PER roleplayers.

Again, the answer is NO. Bargnani has been playing against starters for the past 1.5 years. The ON/OFF DRTG measures the team defence for those 1.5 years + 2.5 years as a backup.

[Indeed]

Now, you're getting that I don't think the on/off court data is all that helpful in evaluating an individual's data, right? Ultimately you're measuring not the individual himself, but peripheries. You're measuring the team with vs. the team without such a player, in which case that player is only 1/5 of what's being measured. The laugh test results don't always look good, either, like when Tim Duncan or Luke Mbah a Moute score very pedestrianly with it.

The superiority of PDSS to me, is that it's usable for a coach. Dean Oliver, of course, was a college assistant coach himself so his methods are designed especially with coaching decisions in mind. As a coach, I can't really do much with On/Off court 82games.com data. It's helpful to know, but I don't know after reading it some key information, like, "who's giving up what, versus who's stopping what?" All we can do is kind of guess with this method who is doing what, and that's not nearly as helpful IMO. And it's not at all helpful in the case of Calderon, whose numbers changed entirely based on coming off the bench vs. starting.

I know what team DRTG is measuring, but when 4000+ minutes of data with Bargnani and another 4000+ minutes without Bargnani show 2 completely different outcomes, with the same teammates, the same system, and the same opponent, there's only one conclusion to make. I am not using it haphazardly here.

You told me that Jack was better than Calderon defensively according to PDSS, yet we got destroyed worse than with Calderon, when Jack played with Bargnani.

I am probably not going to convince you otherwise, and that's fine. Let's agree to disagree.

I think what Boris means is, DRtg is a 5 man based formula, which has problem on some players even their man are not involving in the actual offense. Eg. You are guarding a corner 3 man, and he never touches the ball (hint: it becomes 4 on 4), then your DRtg is meaningless.

Meanwhile Dean Oliver's formula is individual based, which exactly resolve the above situation, where you are neither a good nor bad defender, because you never contribute on the defensive end.

[supersub15]

I think what Boris means is, DRtg is a 5 man based formula, which has problem on some players even their man are not involving in the actual offense. Eg. You are guarding a corner 3 man, and he never touches the ball (hint: it becomes 4 on 4), then your DRtg is meaningless.

Meanwhile Dean Oliver's formula is individual based, which exactly resolve the above situation, where you are neither a good nor bad defender, because you never contribute on the defensive end.

Yeah, I know what he's trying to say, but I have made my reservations about PDSS. And I have already explained that 9000+ minutes of data is a large enough sample for us to make solid conclusions about Bargnani's contribution on the defensive end. It might not be him directly responsible for all of the +8 points differential in DRTG, but something happens to team defence when he plays that doesn't happen when some other big takes his place in the lineup.

The secondary discussion at this point, as Local_NG_Idiot has pointed out, is whether the bigs are responsible or the perimeter guys.

[Ripp]

Well, firstly we need to remember that this is not some sort of Holy Grail one-number metric. There are constituent parts to PDSS ratings: the stop%, DPoss% and finally the DRat. And yes, it does reflect what actually happened on the floor because at some point you have to reconcile individual play with team performance defensively.

No, it doesn't reconcile with what happens on the floor. Look at the formula for the way DRat is calculated that you posted:

Code: Select all

DRat = TeamDRat + %DPoss x (100 x (Opponent Points / Scoring Possession) x (1 - Stop%) - TeamDRat)


In other words, you are computing DRat by taking some team defensive rating and then adding some number that is specific to the player.
So how exactly do I go from each of those DRats to then figure out what the team's DRat for the season will be? Or what our defense will be for the next game? Or how a particular lineup will perform?

In other words, how do you "reconcile individual play with team play defensively", as you said must be done?

It is very easy to do this with On/Off or even better, APM based tools. However, it appears to not necessarily work well with these Oliver ratings:
http://hoyaprospectus.blogspot.com/2008 ... tings.html
http://sonicscentral.com/apbrmetrics/vi ... 42fad1ee78

So I ask again, how do I go from this individual defensive number to an estimate of how a particular lineup will perform, for example?

I wouldn't say this method is a "surrogate" for PBP metrics, because PBP metrics don't contain defensive data. You don't know who did what, other than if a block, steal, foul or defensive rebound takes place. Secondly, it precedes PBP metrics so it obviously isn't a surrogate anything.

Well, OK. Let me pause asking this question for now until you've answered the one above, since this part of my question depends on the previous part. This is mostly semantics...we can discuss it further later.

Yep, I understood this. And I took issue with the FFTM variable being taken as a positive, in the posts I made on page 25. Makes no sense to give a player credit for sending someone to the FT line, even if they miss...

Sure, it makes sense: often, you choose to foul to stop an easy score to save your team some points. That's a partial stop being recorded on a possession, so why wouldn't it? Secondly, it doesn't count for much so it's not a big issue. Thirdly, it has to be accounted for at some point.

WTF? So if I am a foul-prone guy who sends an opponent to the line on every possession, I am in fact NOT penalized for this in any way? The possession has not ended once I've sent the guy to the line, it is still effectively continuing....
Look, imagine Bizarro Amir Johnson, who fouls on every possession and sends his guy to the FT line, which his guy then hits both of. According to your formula, his Stop Percentage has INCREASED! Do you not see how this issue causes your formula to be divorced from reality?

Again, why am I getting credit for sending someone to the free throw line? Why is that good?

Propose an alternative, then.

You cannot do this, man. That is not the way you defend a technique. If I find a flaw in your technique, the burden is not on me to propose an alternative, but on you to prove the correctness of your technique (or justify it in some way.)

Yes, I saw this too, and the formula listed in Oliver's book. But the explanation doesn't seem very convincing. I mean, think about it yourself...what is the natural way to trade off between defensive rebounds and forced misses? There isn't any. And as you'll notice in the link I posted on page 25 frm Oliver's website (http://www.rawbw.com/~deano/articles/JordanvsOlaj.html), when he first did this he arbitrarily decided to weight defensive rebounds by 1/2.

BoP was written about seven years after that article. Oliver clearly worked with this somewhat and came across a more dynamic and accurate method of estimating the value of a forced miss. Improving the method condemns it, how exactly?

I'm not condemning him for changing the formula. I'm saying that what is the justification for either of them? Instead of 1/2, why not 1/4, 1/8, whatever? Which leads to a better stop percentage formula? Or rather, how do we compare different stop percentage formulas and find the weighting that is most consistent with reality?

I wasn't really concerned too much about this particular aspect of the formula...the value being multiplied against %DPoss was the more interesting variable to me, not really the stuff before it.

That doesn't really apply to PDSS numbers, though.

I guess I should hold off on this comment until you've had a chance to respond to the first point.

Again, neither method is "clearly" superior to anything. Even If your estimate of stop percentage measures reality, then it still isn't clear how good the model is, as Oliver said on his website:

The defensive formulas are much simpler, because methods for defensive evaluation aren't as well developed theoretically. Simply put, I have struggled with evaluating players' defenses for many years now. I came up with a basic method three or four years ago called defensive stops, which are a way of estimating how many times a player stops the opposition from scoring. It's not a bad first approximation, but it misses out on players like Joe Dumars and Glenn Rivers, who prevent their assignments from scoring by not allowing them good looks at the basket. They don't get many defensive rebounds or blocks and don't steal the ball much, but they shut down their men. Doug Steele came up with a good way for accounting for this type of player, the kind of player I call a good man defender. On the other hand, my method does best at evaluating team defense, which would include blocks, steals, and defensive rebounds. Doug has begun including these in his method, too, but he uses a form of linear weights, something I disapprove of rather heartily.

And Dean is referring to the estimated DRtg, there, not PDSS...

My point is that his comment applies to both. Both techniques make strong modeling assumptions that may or may not be correct. But anyway, let's revisit this comment once you've told me how to get from individual ratings to say lineup ratings and team ratings.

No biggie, hope you enjoyed your ribs. But like I said, hopefully you get the chance to look at the concerns I raised with the methodology on page 25. Like I keep saying, all of this stuff that Oliver does is meant to estimate something that we can calculate directly these days. If you don't have time to read the stuff on page 25, at the very least read the section called "Defensive Ratings" from his site:
http://www.rawbw.com/~deano/articles/JordanvsOlaj.html

Anyway, hopefully you see what I'm saying, and why this approach of his has some pretty significant limitations.

I've read his website often, but I don't think you're clear in that he doesn't refer to PDSS on his site at all, and it is much older than Basketball on Paper.

i understand that. But like I said, the problems I raised on page 25 are with PDSS, not the stuff on his site. The only reason I linked to his website is so that everyone here can understand his thought process and how he came up with these formulas, without necessarily owning a copy of his book.

[cdel00]

best set of ballboy posts ever!

Boris has made an excellent statisitically based counter argument with a wealth of insight and analysis.

I respect Sub's findings as more than just anomolous based on the size of the data set. However I suggest that since Bargnani has been used more as a prime time player he has been ON vs other team's best and when he is OFF it's vs the other team's bench therefore when you see the ON/OFF numbers you must consider the nature of the data set as you are comparing 19-25 PER stars vs 10-15 PER roleplayers.

Again, the answer is NO. Bargnani has been playing against starters for the past 1.5 years. The ON/OFF DRTG measures the team defence for those 1.5 years + 2.5 years as a backup.

1629+1861=3490mins
vs
2453+2799=5252mins

You have also shown the horrible defensive slide in the latter years so while the range of data includes both starter and bench the size of the data sets are not equal as he played more mins vs the starters than his replacement did and did so on a far worse defensive team. His years as a backup were his rook and soph season in a new style of game so he had bad numbers in terms of ON/OFF even though he played alot vs bench bigs that negates the positive effect of playing vs bench bigs. Year 3 and 4 Bargnani played way more mins vs other team's starters while doing so on a crap defensive team with an injured PG who was as effective on D as the air he guarded.

I would suggest that Bargnani's ON/OFF numbers are so negative because of the nature of the datasets and as Boris suggests that while Bargs in not the cause of the breakdowns he is still impacted by the effect of the brakedowns. Furthermore I could totally see that the Raps cough up 4-6 points a game more with Bargs ON the court than with him OFF the court due to rebounding but that is a stat for another day

[BorisDK1]

I know what team DRTG is measuring, but when 4000+ minutes of data with Bargnani and another 4000+ minutes without Bargnani show 2 completely different outcomes, with the same teammates, the same system, and the same opponent, there's only one conclusion to make. I am not using it haphazardly here.

You told me that Jack was better than Calderon defensively according to PDSS, yet we got destroyed worse than with Calderon, when Jack played with Bargnani.

I am probably not going to convince you otherwise, and that's fine. Let's agree to disagree.

Jack performed much better defensively than Calderon did. I pointed out what happened when Calderon didn't play. 82games.com cannot tell you 1) who was responsible for how much of what, 2) what caused it, 3) against whom it was caused, which are all necessary for any on/off court counting to be particularly meaningful. You are trying to use those metrics with far greater weight and scope than they're warranted. Contrary to this presentation, your stats do not isolate for one or two or even three players: teams, last I looked, still are comprised of five players on the floor at once. Your tabulations are very, very noisy and really don't deal with any player or combination of players in a direct fashion. That's not to say they're meaningless, just that they are more limited than you'd like to think.

Ultimately, I think the disconnect is that I'm a basketball guy (I'm a coach and former official) who has become conversant in statistical measures and wants to use them for the purpose of basketball teaching and coaching; you're a statistical guy who might not have the background in the game that I do, from a theoretical standpoint. I think there's probably going to be a bit of a disconnect there, neither side being completely right, but I think we have to learn from each other and there is room for a lot of accord.

[Ripp]

You sure it is ONLY you being substitute? Unless they both played the same position and against the same offense and system, otherwise, you have two different formulas here:

i) DRtg1 = x1 + x2 + x3 + x4 + x5
ii) DRtg2 = x1 + x2 + x6 + x4 + x7
=> DRtg1 - DRtg2 = (x3 + x5) - (x6 - x7) --- By substitution law (i and ii)
is not equal to Drtg1 > DRtg2 <=> x5 > x7

That is a great point you raise. But you can still handle this. If you have a system of N linear independent equations, you can resolve up to N variables. In the example you gave, I'm trying to resolve 3 variables (Indeed, Deron Williams and Ripp) but only have two equations. So I'd need at least one more equation.

(To be clear, it isn't quite simple linear algebra...the dataset is a lot more complicated than this, there is noise, bias, etc. But the nice thing about having massive amounts of data is that it washes away the noise, allowing you to conceptually think about the problem in its simplest form.)

[BorisDK1]

And he's attempting to defend this POV with the theory that perimeter defense and ball pressure is more important than interior big men who play defense which is so far out ther I'm surprised no one else has picked up on this.

Pressure on the ball is the foundation of any defense. That's defensive basketball 101 for coaches. It's not so "far out there" as you'd like to think. I think you're taking a general truism (great defensive bigs can far more greatly impact a team's defense than great perimeter players) and taking that to unwarranted places (you cannot be a great defensive team without a great defensive big, or if your team's defense breaks down it's the bigs' fault, or a great defensive big can make up for egregious inability to pressure the basketball).

What do you do if you don't have a dominant interior presence? Quit? Odd that some teams can still compete defensively without one, no? What about the poor defensive teams have dominant interior defenders, yet who cannot pressure or stop the basketball on the perimeter, who still manage to suck?

At some point, the purpose of statistics is to narrow down our inquiry as to what is going on on the basketball floor. Some people here want to do basketball analysis only with some statistics, but obviously aren't going to the floor very much with questions. On/off court analysis should lead to further inquiry, they don't provide the answers of themselves.

[Ripp]

I can't be bothered to read all these posts, but is the crux of Boris's argument that the defensive woes of the Raps were no Bargnani's fault, but someone else? Or is he saying Bargnani is a good or passable C defensively?
.

Boris is arguing that Jose was the biggest reason for our defensive woes based on his stats.

Everyone else is saying Bargs was the prime culprit based on drtg (on vs off).

I made this point in a very long post, so it might get lost. So the fundamental thing behind PDSS is calculating Stop %. If some hypothetical perfect defender (Superman? Green Lantern? Wolverine?) played basketball, he'd have a Stop% of 100%. Jose Calderon on the other hand has a very low Stop%.

Stop % is calculated from the number of "Stops" you got. Here are how stops are calculated:

Code: Select all

Stops = FTO + Stl + (FFT / 10) + (FM + Blk) * Forced Miss Weight * (1 - OppOReb%) + DReb * (1 - Forced Miss Weight)


FFT = Forced Free Throw. In other words, under PDSS a foul-prone player like Amir Johnson actually gets credit for sending people to the FT line! Again, why am I trusting a system that rewards a player for earning easy points at the FT line? A guy like Bargnani who gives up many And1s is also rewarded by this stat!

That is why I have no clue what this method is supposed to measure, or what I'm supposed to infer from it.

[Tony_Montana]

BARGNANI IS AN AMAZING FREE THROW DEFENDER SHUT UP.

[dagger]

At some point, the purpose of statistics is to narrow down our inquiry as to what is going on on the basketball floor. Some people here want to do basketball analysis only with some statistics, but obviously aren't going to the floor very much with questions. On/off court analysis should lead to further inquiry, they don't provide the answers of themselves.

By now, Boris, you probably realize that there are some here who have formed a religion based on Bargnani as the root of ALL evil, and you are daring to shake their orthodoxy by presuming that he is only part of our defensive problems. They will resort to a world-is-square argument before admitted that perhaps your conclusions require further scrutiny.

[Kabookalu]

Why the **** is this thread still going on?

[BorisDK1]

No, it doesn't reconcile with what happens on the floor. Look at the formula for the way DRat is calculated that you posted:

Code: Select all

DRat = TeamDRat + %DPoss x (100 x (Opponent Points / Scoring Possession) x (1 - Stop%) - TeamDRat)


In other words, you are computing DRat by taking some team defensive rating and then adding some number that is specific to the player.
So how exactly do I go from each of those DRats to then figure out what the team's DRat for the season will be? Or what our defense will be for the next game? Or how a particular lineup will perform?

In other words, how do you "reconcile individual play with team play defensively", as you said must be done?

How do you project anything with on/off court data? At least part of that tool is based on who's replacing you, and only partially measures (and indirectly!) what you're doing. Can I project performance with the DRat from PDSS? Sure. I can see who's failing more vs. who's succeeding more, I can see what aspects of whose game needs work, etc.

Oliver's justification for his formulae (which took him a lot more time in experimentation, research and development and peer review than your brief critique, for one) is that at some point you cannot talk about an individual's defense without measuring how the team performed, that you have to reconcile them both at some way. It's ridiculous to speak of Amir Johnson as having a great defensive year when he's on a horrid defensive team: at some point, he was unable to singly prevent the badness and is partially responsible. That is a reasonable point, one that you're not interacting with at all.

It is very easy to do this with On/Off or even better, APM based tools. However, it appears to not necessarily work well with these Oliver ratings:
http://hoyaprospectus.blogspot.com/2008 ... tings.html
http://sonicscentral.com/apbrmetrics/vi ... 42fad1ee78

So I ask again, how do I go from this individual defensive number to an estimate of how a particular lineup will perform, for example?

There are a lot of methodological errors in the first article. Firstly, again, that's using the estimated system, not PDSS. Secondly, usage was not factored in with his use of the offensive rating - you cannot isolate efficiency from usage.

The second one was answered sufficiently well by Neil Paine, which of course nobody responded to.

WTF? So if I am a foul-prone guy who sends an opponent to the line on every possession, I am in fact NOT penalized for this in any way? The possession has not ended once I've sent the guy to the line, it is still effectively continuing....

I think you missed something again - maybe in extreme excitement for being able to use "WTF?" to starting a paragraph.

You are penalized if players make their free throws, as you should be. You are creating scores for the opponent in that case. But if guys are going to the line and missing, you are not creating scores so how can you be charged for allowing them, as though you were? They aren't scoring if they're missing, are they?

Look, imagine Bizarro Amir Johnson, who fouls on every possession and sends his guy to the FT line, which his guy then hits both of. According to your formula, his Stop Percentage has INCREASED! Do you not see how this issue causes your formula to be divorced from reality?

You whiffed on that one. FFT count towards scoring possessions, not stops.

FFT = forced free throws
FFTM = forced free throws missed

You cannot do this, man. That is not the way you defend a technique. If I find a flaw in your technique, the burden is not on me to propose an alternative, but on you to prove the correctness of your technique (or justify it in some way.)

Maybe it would help if what you call a "flaw" were either 1) understood, 2) represented correctly, 3) an actual flaw.

I'm not condemning him for changing the formula. I'm saying that what is the justification for either of them? Instead of 1/2, why not 1/4, 1/8, whatever? Which leads to a better stop percentage formula? Or rather, how do we compare different stop percentage formulas and find the weighting that is most consistent with reality?

The FMwt formula assigns the value on what the team is actually doing, not on some absolute number divorced from reality in that game/series of games/season. Oliver justifies this by pointing to high school or college teams who might play zone and hold opponents' shooting really low, but give up a ton of offensive rebounds. The value of that team's forced misses are far lower than a team who gives up a lot of scores but controls the defensive board.

My point is that his comment applies to both. Both techniques make strong modeling assumptions that may or may not be correct. But anyway, let's revisit this comment once you've told me how to get from individual ratings to say lineup ratings and team ratings.

No, it really doesn't apply to both in the same way. The first method is an estimate, the latter is far, far more exact.

[BorisDK1]

I can't be bothered to read all these posts, but is the crux of Boris's argument that the defensive woes of the Raps were no Bargnani's fault, but someone else? Or is he saying Bargnani is a good or passable C defensively?
.

Boris is arguing that Jose was the biggest reason for our defensive woes based on his stats.

Everyone else is saying Bargs was the prime culprit based on drtg (on vs off).

I made this point in a very long post, so it might get lost. So the fundamental thing behind PDSS is calculating Stop %. If some hypothetical perfect defender (Superman? Green Lantern? Wolverine?) played basketball, he'd have a Stop% of 100%. Jose Calderon on the other hand has a very low Stop%.

Stop % is calculated from the number of "Stops" you got. Here are how stops are calculated:

Code: Select all

Stops = FTO + Stl + (FFT / 10) + (FM + Blk) * Forced Miss Weight * (1 - OppOReb%) + DReb * (1 - Forced Miss Weight)


FFT = Forced Free Throw. In other words, under PDSS a foul-prone player like Amir Johnson actually gets credit for sending people to the FT line! Again, why am I trusting a system that rewards a player for earning easy points at the FT line? A guy like Bargnani who gives up many And1s is also rewarded by this stat!

That is why I have no clue what this method is supposed to measure, or what I'm supposed to infer from it.

My apologies.

That should be "FFTM" - forced free throw missed. My bad, I caused that confusion.

[Courtside]

The secondary discussion at this point, as Local_NG_Idiot has pointed out, is whether the bigs are responsible or the perimeter guys.

I don't think it's an either/or. Pretty much anyone can see it's a measure of both, so the debate is more about which was the larger issue on the Raptors and what can be done to address it.

Looking at Boston, is Perkins really that good a help defender, or did the ability of the wing defenders to to slow down their man by a half step give Kendrick the extra half step he needed to get there? Put him behind Calderon, DeRozan, Turkoglu and Bosh - is he anywhere near as effective? Probably not.

Clearly Bargs is not as defensively as aware as he needs to be and is often that half step slow getting to where he needs to be. So we know he's already weak in this regard, but if he played with competent defenders at the other 4 spots, would he still be as bad? Again, probably not.

Dwight Howard wouldn't have been able to cover up the disaster we had from the other 4 positions last year. An already weak defender was being asked to erase what was likely the largest number of blow by's and lapses in the league, so if course he's going to stand out, and of course the better defender that is his backup is going to look better playing against other backups.

The question is - where does it go from here?

[cdel00]

Amir sends the right people to the line instead of giving up a free lane and the chance for an and1. That actually is helpful when you play with soft weak side help. I give Amir credit when he sends a guy who just blew by Jose and was about to cough up a rebound dunk on Bargnani's head to the line where he only makes 1 of 2 that is something that deserves credit cause that's called defensive intimidation

[Dam]

At some point, the purpose of statistics is to narrow down our inquiry as to what is going on on the basketball floor. Some people here want to do basketball analysis only with some statistics, but obviously aren't going to the floor very much with questions. On/off court analysis should lead to further inquiry, they don't provide the answers of themselves.

By now, Boris, you probably realize that there are some here who have formed a religion based on Bargnani as the root of ALL evil, and you are daring to shake their orthodoxy by presuming that he is only part of our defensive problems. They will resort to a world-is-square argument before admitted that perhaps your conclusions require further scrutiny.

It's obvious..
Boris is a waste of time

[Indeed]

At some point, the purpose of statistics is to narrow down our inquiry as to what is going on on the basketball floor. Some people here want to do basketball analysis only with some statistics, but obviously aren't going to the floor very much with questions. On/off court analysis should lead to further inquiry, they don't provide the answers of themselves.

By now, Boris, you probably realize that there are some here who have formed a religion based on Bargnani as the root of ALL evil, and you are daring to shake their orthodoxy by presuming that he is only part of our defensive problems. They will resort to a world-is-square argument before admitted that perhaps your conclusions require further scrutiny.

I am guessing Boris has been following along on RealGM for a while, and would really like to get involved in this thread.

[OneTime86]

Why the **** is this thread still going on?

+1

I never really understood the point of it in the first place.

[Indeed]

Why the **** is this thread still going on?

+1

I never really understood the point of it in the first place.

Someone introduced another way to determine a player's defensive ability in terms of stats.
Perhaps there are questions on why it is a more accurate, and how effective/flaw stats can be to proof a point.