Is PDSS a good way to analyze the Raptors defense?

[Laowai]

I like the debate and like statistics as much as anyone but the unfortunate thing is that statistics take out the human factor and the lineups faced in each game and each situation.
I think the merits of statistics tend to get lost because it is virtually impossible to track each game precisely taking into consideration who the opposition at the times was.
Other factors that can't be judged easily by statistics is when a player is a peak efficency in a game Whether minutes played effect the proficentcy of defense or if a player plays an extended stretch does the efficentcy rating go down or up. Defense's in real life are effected by the opposition does player A match up better with player B in a particular game.

While learning from history is important we are actually debating the past and not the present.
The statistical imperatives that defined last year no longer define this year. Just like looking back at 2007-2008 when we were much better defensively relate to todays roster.

Unless a model can be produced to simulate the interaction of new players and how the loss of other players significantly change the dynamics of the model.

[raptorforlife88]

Ok I think a lot of people reading this thread are just skimming through it, so I think it should be made clear that Boris watched every single game and every single play last year for the Raptors, tracking each possession for PDSS. It's posted on Raptorspace forums.

Each individual one is available in all 82 game threads. This isn't something that's coming just from numbers from 82games or something. Boris watched every single freaking play.

[Scott Carefoot]

It's pretty extraordinary the lengths people seem to be willing to go to make their case that Bargnani or Calderon is more to blame for the Raptors' terrible defence. The fact of the matter is that they're both terrible. Calderon should be traded or should be a change-of-pace backup and Bargnani should be a sixth man until he proves that he can rebound or defend adequately. The latter isn't going to happen, of course, which means this debate will rage on indefinitely since the Raptors' defence will continue to be porous.

[BorisDK1]

I haven't looked at the formulas but as an illustration of how a small sample can skew things lets say players Alex and Paul are great one on one defenders. The offense however decides to go with pick and roll and Alex/Paul end up giving up layups 90% of the time since they don't know how to defend the pick and roll. In this case the team will look bad but the individual drtg of Alex and Paul will look good.

If Alex and Paul are put in a lot of PnR situations and they keep getting beaten, they're going to surrender a lot of FGA and not produce many stops. In that case, their stop% is going to look weak which will make their DRat look fairly weak.

How the team's defense would look would depend upon how well they defend everything else. When we are speaking of "team" in the PDSS sense, we're speaking solely of a phenomenon in which a player scores without being guarded by a specific individual. I'll run through a list of some examples:

• In a 2-on-1, the defender stops the ballhandler but the pass results in a layup for his teammate. One guy can't be held accountable for guarding two: that's a "team" FGA;
• Help-and-recover situations, like doubling down to the post with no rotation to cover the perimeter results in the outcome going to "team", again, because one person cannot be held accountable for two and the team is making a conscientious decision to allow the lesser of two evils (in their view). The only exception for that is if a defensive player makes a herculean effort to get sprinting on air time to close out in time to visibly bother the shot: I'll reward that effort with a FM for that defender;
• A pass to a player under no pressure from the defense steps out of bounds. That FTO goes to "team";
• Complete breakdowns of the defense where it's just impossible, even upon multiple reviews, to determine who specifically was at fault or even narrow it down to two guys. Some offensive rebounding situations could easily be attributable to three or four guys and at that point you just have to give that to "team";
• 3-, 5-, 8- and 24-second violations are attributed to "team".

[Ripp]

3) If I have a team with team Drtg of say 113, and say only one player with an individual DRat higher than this, is it safe to say that he is the worst defender on the team? Should it not then be true that lineups that don't have this guy have Drtgs better than the team average of 113?

Not necessarily, because the entirety of team play isn't encapsulated in one individual's defensive performance.

It take it that the "Not necessarily" is in reference to my second question, right? If so, you are saying that my team defense will not necessarily improve if I remove the worst defenders (as ranked by PDSS Drtg)? I'm not trying to mis-attribute stuff to you, just trying to understand what you are saying.

[BorisDK1]

Did you not understand my point about averages? If you remove the biggest number from a list, then the average must decrease. Remove the smallest, average has to go up. Similarly, if Jose is the worst defender and by far the worst defender, then I'd better see lineups w/o him doing pretty well...

And what happens when you see the team with Jose in civilian clothes? *gasp* They get a lot better...

[timdunkit]

Here is my problem with these stats ... how can you apply them specifically? My problem with a stat like DRTG has always been that it done by taking 100 possessions. The problem is that, no team (maybe GSW?) plays at over that pace and no player plays for 48 minutes. The normal starter playing 36 minutes and whose teams play at an average 93 possession a game would see 70 possessions a game. If a DRTG difference is about 4 points, then the actually game difference is about 70% of that which is 2.8 points. When people put DRTG numbers up, they put them at 100 paces yet thats not a true reflection of whats happening in the game.

In the other thread, SS posted the DRTG of Bargs/Jack and no Bargs/Jack. The difference was 4 points but when you accounted the Jack minutes (using him as the limited reagent) the maximum in game difference was just one basket and that would have been if Jack didn't play Bargs at all. If you were a coach, and your start guy told you that the difference between games of playing this guy without this guy, would leave to a one shot improvement in your defense ... well hes going to laugh at you. Its no wonder Triano said they don't use DRTG because these stats don't have applications. They can give you a general idea of how a team is defensivey, but you can't conclude anything specific to it.

Think about the error involved in the calculations. DRTG and Pace are averages over an 82 games ... yet no one at this point has bothered to calculate the error calculations that comes with it or the frequency at which the Pace or DRTG is between two points.

Lastly, the problem with defensive statistic is that there all based on one flaw that makes it total unrealiable ... even PDSS. The idea behind all the defensive statistics we have is that if we measure what the offense is doing against a certain defense, then we can know how that defense is doing. We AREN'T measuring defense directly ... we are trying to measure it indirectly by measuring a teams offense. Which makes sense except when you realize that most defenses have become dynamic and require a team in sync with each other from top to bottom.

What defensive statistic needs is a evolution and a better thought process. The only time I think its okay to measure what an opponent is doing against a defense to give you an idea of the defense is in isolation situations, which basically comes down to 1 on 1 play. Most defenses need something more advance because saying what the offense does against a certain defense, doesn't tell us what that defense is doing. Defense statistics need to be altered and become focused on whats happening on defense. They shouldn't be based on ppg or assists or w/e, because those are offensive statistics. Things like, which guards are the best at fighting through screens, and which get caught up on picks along with the frequency that the guard gets caught up is a defensive statistics. Whats the avg shots that a big man challenges in the paint is a defensive statistics. You could even get more complex (if you had the time to watch every PnR) to figure out which big man are the strongest hedges on PnR, by calculated the time that the big is in hedge motion and the frequency ...


Defense is a little more complicated then offense when it comes statistics but if we really want to get stats that we can apply then we need to focus on creating defensive statistics and not using offensive statistics that aren't designed to measure defense ...

[bthrawn]

Boris in the other thread you said you didn't track 3 pointers correctly. Was that just in terms of not having a column to track 3 point versus 2 point FGs? Or was it also in how you assigned points allowed? Also in terms of your stop % did you track 3 pointers as 2 pointers? This could result in why your Drtg /= 113.

[Ripp]

Did you not understand my point about averages? If you remove the biggest number from a list, then the average must decrease. Remove the smallest, average has to go up. Similarly, if Jose is the worst defender and by far the worst defender, then I'd better see lineups w/o him doing pretty well...

And what happens when you see the team with Jose in civilian clothes? *gasp* They get a lot better...

Indeed, this is true. And your PDSS system ranked Andrea Bargnani as an above average defender relative to the team. In fact, he and Bosh were tied for the 5th best defense on the team, after Pops Mensah-Bonsu, Evans, Banks, and Amir. His individual Drtg is was 110, and the team Drtg 113.2.

Yet as supersub15 showed, when Bargnani is on the bench or otherwise not playing, the Raptors defense is substantially better.

Why is the defense improving dramatically when one of the team's best defenders (as ranked by PDSS) is off the floor?

[BrickHeads]

Did you not understand my point about averages? If you remove the biggest number from a list, then the average must decrease. Remove the smallest, average has to go up. Similarly, if Jose is the worst defender and by far the worst defender, then I'd better see lineups w/o him doing pretty well...

And what happens when you see the team with Jose in civilian clothes? *gasp* They get a lot better...

Indeed, this is true. And your PDSS system ranked Andrea Bargnani as an above average defender relative to the team. In fact, he and Bosh were tied for the 5th best defense on the team, after Pops Mensah-Bonsu, Evans, Banks, and Amir. His individual Drtg is was 110, and the team Drtg 113.2.

Yet as supersub15 showed, when Bargnani is on the bench or otherwise not playing, the Raptors defense is substantially better.

Why is the defense improving dramatically when one of the team's best defenders (as ranked by PDSS) is off the floor?

Correct me if I'm wrong, but Amir did come in to relieve Bargnani right? He was also rated a better defender no? Could also be the fact that Bargnani played against starters, and his subs played against subs?

[Ripp]

Here is my problem with these stats ... how can you apply them specifically? My problem with a stat like DRTG has always been that it done by taking 100 possessions. The problem is that, no team (maybe GSW?) plays at over that pace and no player plays for 48 minutes. The normal starter playing 36 minutes and whose teams play at an average 93 possession a game would see 70 possessions a game. If a DRTG difference is about 4 points, then the actually game difference is about 70% of that which is 2.8 points. When people put DRTG numbers up, they put them at 100 paces yet thats not a true reflection of whats happening in the game.

Good question. If your DRtg is say 115, it means that on average, opposing teams scored 1.15 points on you for every possession. So if you want to answer the question, "on average, how many points are going to be scored on my team in 10 possessions?", Drtg is a pretty good way of estimating it.

That is all it really says..how many points does the other team score each time the go up the floor, on average.

In the other thread, SS posted the DRTG of Bargs/Jack and no Bargs/Jack. The difference was 4 points but when you accounted the Jack minutes (using him as the limited reagent) the maximum in game difference was just one basket and that would have been if Jack didn't play Bargs at all. If you were a coach, and your start guy told you that the difference between games of playing this guy without this guy, would leave to a one shot improvement in your defense ... well hes going to laugh at you. Its no wonder Triano said they don't use DRTG because these stats don't have applications. They can give you a general idea of how a team is defensivey, but you can't conclude anything specific to it.

The average (and median too, I believe) margin of victory in most seasons by the home team is between 3 or 4 points. Thus, basketball games are damn close, so even a "one shot improvement" is quite frequently the difference between winning and losing. Basically, it makes sense to do the best you possibly can, rather than just thinking the possessions don't matter.

[bthrawn]

Did you not understand my point about averages? If you remove the biggest number from a list, then the average must decrease. Remove the smallest, average has to go up. Similarly, if Jose is the worst defender and by far the worst defender, then I'd better see lineups w/o him doing pretty well...

And what happens when you see the team with Jose in civilian clothes? *gasp* They get a lot better...

Indeed, this is true. And your PDSS system ranked Andrea Bargnani as an above average defender relative to the team. In fact, he and Bosh were tied for the 5th best defense on the team, after Pops Mensah-Bonsu, Evans, Banks, and Amir. His individual Drtg is was 110, and the team Drtg 113.2.

Yet as supersub15 showed, when Bargnani is on the bench or otherwise not playing, the Raptors defense is substantially better.

Why is the defense improving dramatically when one of the team's best defenders (as ranked by PDSS) is off the floor?

You could also say why is the Raptors defense better when their leading shot blocker isn't playing. He lead the team by a ton over Bosh/Amir. This indicates that there are some things bargs is doing well at the defensive end but something else he's doing poorly which throws the numbers out.

The following is a hypothesis which I don't know how to prove.

Why not look at # of possessions per minute/36 that Barg's faces compared with say Amir. Bargs is a bad rebounder specially on the offensive end which could lead to 1-2 extra defensive possessions faced per game which is actually why team defense improves with him off the court. They just face fewer shots. But he could also be an average overall defender but looks worse in terms to other players due to facing more shots. All of the above could be true. Average defender facing more shots than another average defender is going to look worse when compared.

[ranger001]

I haven't looked at the formulas but as an illustration of how a small sample can skew things lets say players Alex and Paul are great one on one defenders. The offense however decides to go with pick and roll and Alex/Paul end up giving up layups 90% of the time since they don't know how to defend the pick and roll. In this case the team will look bad but the individual drtg of Alex and Paul will look good.

If Alex and Paul are put in a lot of PnR situations and they keep getting beaten, they're going to surrender a lot of FGA and not produce many stops. In that case, their stop% is going to look weak which will make their DRat look fairly weak.

Ok I just picked PNR as an example but my point is that if the team fails a lot of times then it will affect the team drtg a lot without affecting the individual drtg as much.

[Reignman]

Here is my problem with these stats ... how can you apply them specifically? My problem with a stat like DRTG has always been that it done by taking 100 possessions. The problem is that, no team (maybe GSW?) plays at over that pace and no player plays for 48 minutes. The normal starter playing 36 minutes and whose teams play at an average 93 possession a game would see 70 possessions a game. If a DRTG difference is about 4 points, then the actually game difference is about 70% of that which is 2.8 points. When people put DRTG numbers up, they put them at 100 paces yet thats not a true reflection of whats happening in the game.

In the other thread, SS posted the DRTG of Bargs/Jack and no Bargs/Jack. The difference was 4 points but when you accounted the Jack minutes (using him as the limited reagent) the maximum in game difference was just one basket and that would have been if Jack didn't play Bargs at all. If you were a coach, and your start guy told you that the difference between games of playing this guy without this guy, would leave to a one shot improvement in your defense ... well hes going to laugh at you. Its no wonder Triano said they don't use DRTG because these stats don't have applications. They can give you a general idea of how a team is defensivey, but you can't conclude anything specific to it.

Think about the error involved in the calculations. DRTG and Pace are averages over an 82 games ... yet no one at this point has bothered to calculate the error calculations that comes with it or the frequency at which the Pace or DRTG is between two points.

Lastly, the problem with defensive statistic is that there all based on one flaw that makes it total unrealiable ... even PDSS. The idea behind all the defensive statistics we have is that if we measure what the offense is doing against a certain defense, then we can know how that defense is doing. We AREN'T measuring defense directly ... we are trying to measure it indirectly by measuring a teams offense. Which makes sense except when you realize that most defenses have become dynamic and require a team in sync with each other from top to bottom.

What defensive statistic needs is a evolution and a better thought process. The only time I think its okay to measure what an opponent is doing against a defense to give you an idea of the defense is in isolation situations, which basically comes down to 1 on 1 play. Most defenses need something more advance because saying what the offense does against a certain defense, doesn't tell us what that defense is doing. Defense statistics need to be altered and become focused on whats happening on defense. They shouldn't be based on ppg or assists or w/e, because those are offensive statistics. Things like, which guards are the best at fighting through screens, and which get caught up on picks along with the frequency that the guard gets caught up is a defensive statistics. Whats the avg shots that a big man challenges in the paint is a defensive statistics. You could even get more complex (if you had the time to watch every PnR) to figure out which big man are the strongest hedges on PnR, by calculated the time that the big is in hedge motion and the frequency ...


Defense is a little more complicated then offense when it comes statistics but if we really want to get stats that we can apply then we need to focus on creating defensive statistics and not using offensive statistics that aren't designed to measure defense ...

Yeah, I agree with this completely. PDSS and other stats leave a lot to desire as soon as other factors get thrown in the mix. It seems like it would be a great tool in 1 vs 1 situations when the "help" is taken out of the equation. Unfortunately "help" is what dictates how good or bad your defense will be in the NBA.

[supersub15]

It's pretty extraordinary the lengths people seem to be willing to go to make their case that Bargnani or Calderon is more to blame for the Raptors' terrible defence. The fact of the matter is that they're both terrible. Calderon should be traded or should be a change-of-pace backup and Bargnani should be a sixth man until he proves that he can rebound or defend adequately. The latter isn't going to happen, of course, which means this debate will rage on indefinitely since the Raptors' defence will continue to be porous.

The discussion has turned into who's more responsible for the breakdown in defence, the PG or the big? Boris is saying it's the PG, a lot are saying the big. Boris is using PDSS, I'm using DRTG ON/OFF.

If Calderon is the main reason for our defensive breakdowns last year, how come the team DRTG was only 104.02 last year with Calderon and 2 bigs other than Bargnani (for around 500 minutes, including the dreaded Calderon-Jack combo)? Boris has called it coincidental correlation (Post hoc ergo propter hoc), but when the same thing happens every year, since 2007, it stops being coincidental. Our defence doesn't suddenly play that much better just because.

[Ripp]

Correct me if I'm wrong, but Amir did come in to relieve Bargnani right? He was also rated a better defender no? Could also be the fact that Bargnani played against starters, and his subs played against subs?

This is a good point you raise...the defense could certainly improve if say the non-Bargnani minutes were filled in by superior defenders. However, the question is does the difference in the quality of the guys replacing him justify the 9 point swing in on/off numbers?
Heck, Amir is at 109.5, Bargs at 110...does that .5 swing in PDSS Drtg represent a 9 point swing in actual lineup play?
Regarding bias,...it doesn't seem plausible that the numbers were biased against bench players when he was a bench guy, and then biased against starters when he started.

[ranger001]

The discussion has turned into who's more responsible for the breakdown in defence, the PG or the big? Boris is saying it's the PG, a lot are saying the big. Boris is using PDSS, I'm using DRTG ON/OFF.

If Calderon is the main reason for our defensive breakdowns last year, how come the team DRTG was only 104.02 last year with Calderon and 2 bigs other than Bargnani (for around 500 minutes, including the dreaded Calderon-Jack combo)? Boris has called it coincidental correlation (Post hoc ergo propter hoc), but when the same thing happens every year, since 2007, it stops being coincidental. Our defence doesn't suddenly play that much better just because.

The thing with coincidental correlation though is that you can't create a linkage just because they always follow each other. e.g. Just because roosters always crow before the sun comes up doesn't mean they make the sun come up after you've observed this behaviour for years.

I understand your example is better linked than that just saying you advance the linkage differently than by saying 'its happened for years'.

[timdunkit]

Good question. If your DRtg is say 115, it means that on average, opposing teams scored 1.15 points on you for every possession. So if you want to answer the question, "on average, how many points are going to be scored on my team in 10 possessions?", Drtg is a pretty good way of estimating it.

That is all it really says..how many points does the other team score each time the go up the floor, on average.

Here is the thing Ripp, if a DRTG is 110 vs 112, then it means that the latter guy who is seen worse defensively is onyl allowing 0.02 pts more defensively ever possession ... thats not applicable to a game scenerio.

The average (and median too, I believe) margin of victory in most seasons by the home team is between 3 or 4 points. Thus, basketball games are damn close, so even a "one shot improvement" is quite frequently the difference between winning and losing. Basically, it makes sense to do the best you possibly can, rather than just thinking the possessions don't matter.

A one shot improvement can come from just about anywhere ... and when you take the OPP field goals made, anyone of those shots could have been better played defensively in someway to cause a miss.

Like I siad in my long post, defensive stats we have now aren't applicable to defense because they measure what an OPP offense is doing and not actually what the defense is doing.

Edit: Also like to add, that one shot could be a free throw difference when a player isn't even playing defense. Take into account that the Raptors OPP shot the highest free throw% across the league.

[disoblige]

- All of our players often could not stop their man.
- We get out rebounded and we gave extra possessions.
- Bosh and Bargnani did not put much resistant in the paint. It's their job.
- Too much turnovers after the all-star break.

For me Bargs, Jose , Derozan and Hedo were the major reason we sucked defensively. It's remarkable we were once 4th-5th in the east despite that we were last defensively and we did not start the season well.

[supersub15]

The discussion has turned into who's more responsible for the breakdown in defence, the PG or the big? Boris is saying it's the PG, a lot are saying the big. Boris is using PDSS, I'm using DRTG ON/OFF.

If Calderon is the main reason for our defensive breakdowns last year, how come the team DRTG was only 104.02 last year with Calderon and 2 bigs other than Bargnani (for around 500 minutes, including the dreaded Calderon-Jack combo)? Boris has called it coincidental correlation (Post hoc ergo propter hoc), but when the same thing happens every year, since 2007, it stops being coincidental. Our defence doesn't suddenly play that much better just because.

The thing with coincidental correlation though is that you can't create a linkage just because they always follow each other. e.g. Just because roosters always crow before the sun comes up doesn't mean they make the sun come up after you've observed this behaviour for years.

I understand your example is better linked than that just saying you advance the linkage differently than by saying 'its happened for years'.

You're using the wrong analogy here. I'm not making assumptions based on seeing the rooster crow. For 4 full years and thousands of minutes, I've measured the distance from rooster to the sun (lol), chronicled the time of day that the rooster crows, chronicled the time of day that the sun comes up, then established scientific correlation.

Again, if Calderon is such a terrible defender, why our OTHER bigs are capable of covering for him. And don't give me the starter/bench argument. Calderon has played against both, with and without Bargnani.