Talk Me Down: How is Deng not MVP of the Bulls?

[Rerisen]

Ok. By how much? And how do you explain this?

A lot less. Maybe between 6-15 point victories depending. They murdered a bunch of bad teams.

I was giving you the last word though, not sure why you keep asking questions, that you just want to give your own answers to. Now you are going to say, based on the numbers you love from last year, even if those games were deviations, the Bulls should still have been expected to win them bigger with Deng than with Rose. No? Why are we still beating this dead horse...

More interesting in that stretch though, the Bulls only played 3 even 'decent' teams. Memphis, and Boston twice. These are teams that I would expect a Bulls without Rose team to be close to if he wasn't there a full year (maybe a bit worse). Moving toward the type of teams that I said early in the thread would probably expose the loss of Rose more.

Well their combined margin in those 3 were -11 (Deng -17) Boston one home, one away, Memphis on the road, so maybe shave off 2 or 3 points there. Still down. Seems like they fared worse than we would expect if they were to be a .500 team without Rose, let alone better than .500.

Only 3 games, so even smaller than the 10 game sample. But interesting. You asked how many games I needed, well a lot more vs opponents like this and even better ones (Boston was only .500). Of course that's probably not going to happen this year. So we are at a dead end. Could perhaps revisit the thread later in the year, but these two stretches are likely to overshadow all the forthcoming games where they are together, since they are both such high minute players.

[mysticbb]

A lot less. Maybe between 6-15 point victories depending. They murdered a bunch of bad teams.

Well, "murdering a bunch of bad teams" is what better than average teams are doing, while they are usually struggling against better teams.

You gave out a spread here, assuming you want to go game by game, but as I pointed out, capping specific games to a max scoring margin decreases the overall predictive power.

Well, let us take a look at the opponents. In average those opponents in those 10 games had -5.37 SRS. That is indeed bad, no doubt about that. We also know that the Bulls played 6 games at home and 4 on the road. With a +3.2 HCA that gives us in average +0.64 as expected scoring margin for an average team based on HCA alone. Overall that gives us 6.01 as expected scoring margin for those 10 games, if the Bulls would have been an average team. The scoring margin was 9.4 in favor of the Bulls in average, that gives us 3.39 SRS for those 10 games.
Well, if you say it is 6, we agree, everything beyond that is not backed up by facts. 15 is completely out of the question, because that would mean you would expect the Bulls to be a -5.6 SRS team. Maybe you can give out a more specific average value in order to establish a base for the difference between our opinions.

I was giving you the last word though, not sure why you keep asking questions, that you just want to give your own answers to.

No, I actually really want to understand the differences here. ;)

More interesting in that stretch though, the Bulls only played 3 even 'decent' teams. Memphis, and Boston twice.

The interesting part about that is that two of those 3 games were on the road. So, giving the strength of those two teams while adjusting for HCA, we are getting -3.71 as expected scoring margin, if the Bulls are playing like an average team. Well, the Bulls ended up being at -3.67, basically as expected for an average team.

The Bulls so far in this season have played in average teams which are 3 points worse than that combined 3 games. So, in order to do worse than average recordwise, you have to expect the Bulls to play a tougher schedule somehow, otherwise pointing out just those 3 specific games makes not much sense.

[Rerisen]

The interesting part about that is that two of those 3 games were on the road. So, giving the strength of those two teams while adjusting for HCA, we are getting -3.71 as expected scoring margin, if the Bulls are playing like an average team. Well, the Bulls ended up being at -3.67, basically as expected for an average team.

Not following, go by each game. The Boston home and away should be practically a wash against a .500 Bulls team. As the Bulls should also get a HCA on their game. Leaving Memphis as an expected 11 point loss? (They really lost by 16 but had +5 from the two Boston games).

The Bulls so far in this season have played in average teams which are 3 points worse than that combined 3 games. So, in order to do worse than average recordwise, you have to expect the Bulls to play a tougher schedule somehow, otherwise pointing out just those 3 specific games makes not much sense.

If those 10 games were a 180 in opponent strength, teams with records averaging .600+ win percentage say, I would be so much more impressed if they handled them beyond expectations without Rose. Despite not liking 10 games as a sample size. As quality of opponent I see as a critical factor that Rose's specific talents are needed for, and Deng certainly can't answer, with his game so dependent on others offensively.

The entire narrative of this season against the Bulls, is roughly based on the same concept. That despite that the Bulls have a much higher rated offense this year, nearly as good as Miami (and as good Point Diff - at least till last night), it's going to break down in similar catastrophic fashion like it did against the Heat in the ECF last year. Due to that once an entire elite defense focuses on Rose, the cast's inability to produce offense is going to drown the team, much in the same way that I'm suggesting here would happen even more often without Rose in a full season. Based on last year's ECF, it's hard to say that doesn't have merit.

But to neuter the Bulls - a great passing team- in this way you really do need a top defense. These scrub teams can't do it. With or without Rose.

[mysticbb]

Not following, go by each game.

Me neither. Somehow I have the strength of the opponents multiplied by 3, that makes a 1.76 difference of the expected value. So, in fact it should -1.95 as expectation and not -3.71.

If those 10 games were a 180 in opponent strength, teams with records averaging .600+ win percentage say, I would be so much more impressed if they handled them beyond expectations without Rose.

That would make Rose basically worthless. That makes not much sense to base your opinion on beating expectations by so much while one of the most important players is out. I see some huge discrepancies between your expectations and the reality here. And that is likely were the difference between my and your opinion comes from. You expect the Bulls to beat better teams while Rose is out consistently, while that would be a trait of a contender. Contenders are those teams, which are beating up on bad teams while winning more games than losing against better teams. That's what you are asking the Bulls to do without Rose, they should basically play like a contender. They are not, and the numbers aren't predicting such a thing at all.

And it is not like the Bulls beat expectations during the 7 games without Deng against better teams constantly either. Are you here as critical in terms of overall impact level by Rose as well?

[Dr Positivity]

Great last post drza. I think +/- is a better stat than PER, WS, TS%, WP, etc., or raw ppg/apg/rpg alone, but I do have some skepticism when it's treated like the be all end all. As mentioned a few times, it's a phenomenal coorelation stat, but the conclusions drawn from that in terms of eg. Who's the more valuable player in a vacuum, can only be taken so far. As in an example like this, the Bulls success may coorelate most with Deng, but who's responsible for that coorelation is not necessarily Deng alone

[Rerisen]

And it is not like the Bulls beat expectations during the 7 games without Deng against better teams constantly either. Are you here as critical in terms of overall impact level by Rose as well?

No, I don't think the 10 or the 7 game stretch says that much. And I don't expect the Bulls to beat top teams without Rose, actually they really haven't played any. (They did almost beat Miami without Deng).

I'll try to restate my position once more, I don't expect you will agree, but perhaps at least understand where I"m coming from.

Ok so our tiny 3 game sample is at least in keeping with my position. That the team would be (could be) uniquely vulnerable on offense without Derrick Rose (beyond *existing* established with-Rose values), either to high quality defense or to late game pressure.

We just don’t have much data for this. Rose has been playing in virtually all these situations this season and last. When he’s not in the game, neither are all the opponents best players (first unit), who are generally by far most responsible for those teams being so good. Further, if the Bulls have a couple bad possession in a row with him out, the coach can (and does) put Rose back in the game, such that long negative droughts are prevented before they get very bad. Which is why I don't like using a player's minutes from last year against benches to model vs starting units.

Then when you play crap teams (most of this year's 10 games stretch w/o Rose) they have first units that aren’t much better than better teams 2nd units. So they aren’t testing my hypothesis either.

To go sideways a bit here, Carlos Boozer has gotten ripped on the Bulls board this year for his continual feasting on poorer teams relative to good teams. (This is in contrast to Derrick Rose, who did very well vs top competition - numbers ran by Alucryts last season). This is a stigma that has followed him from Utah where he choked it up vs the size and length of the Lakers in the playoffs.

Now if all the data I gave you for Carlos this year, was his performance vs the weak teams, and asked you to use that data to predict his performance against top teams, the prediction would overestimate what he actually did. To get the full truth of his performance, and true overall ability, and thus to make better future predictions for him, you need those higher quality opponent games.

I see no reason for Boozer's splits not to continue in this fashion either. He is a player that is not good at finding his own offense, must get it through the system, and this is something more easily shut down than individual superstar offensive talent. Same as Rose's teammates, relative to this topic at large. What I'm am looking for is really something akin to what splits could tell us - you wouldn't ideally look at Road games to figure out how good a player is at Home - but the column we need is just about blank for the scenario we are trying to project.

[Rerisen]

What we do know is with the team as constructed, Deng is very very valuable. And it's great he's getting such recognition.

Is he, within the system which Rose is part of, supercharged in some fashion to actually be more valuable than Rose? I'm not ready to go there yet, but it will be interesting thing to keep tracking.

Non Prior-informed RAPM now likes Taj Gibson as much as Deng this year.

[mysticbb]

Ok so our tiny 3 game sample is at least in keeping with my position. That the team would be (could be) uniquely vulnerable on offense without Derrick Rose (beyond *existing* established with-Rose values), either to high quality defense or to late game pressure.

The thing is that I agree with that completely, but that doesn't take away the rest at all. As I pointed out, even assuming the basically worst we still end up with equal strong teams without Rose or without Deng.
The only reasonable explanation I have for your position so far is either you are misjudging the strength of schedule (with a bias that seems to think that without Rose the Bulls have to play more often against the better teams) or your expectations upon the team is not reasonable in order to justify that Deng can have a bigger impact than Rose. Everything else is basically covered, and we agree on that much more than you might realize (or want to accept). At the end it may come down to a difference of maybe 1 to 3 points per game (to put some numbers on it) between our views. I can live with that, but I can't live with argumentations based on strawmen or false assumptions about the underlying methods. ;)

Non Prior-informed RAPM now likes Taj Gibson as much as Deng this year.

While playing way less minutes, that should always be kept in mind. ;)
It is also the case that prior informed still is better at predicting the outcome of future games. Prior is just a fixed value each player is pulled to instead of 0 for the non-prior informed version, just for your information.

[lorak]

Prior is just a fixed value each player is pulled to instead of 0 for the non-prior informed version, just for your information.

What does it mean? Could you explain it using different words or some kind of example?

[mysticbb]

Prior is just a fixed value each player is pulled to instead of 0 for the non-prior informed version, just for your information.

What does it mean? Could you explain it using different words or some kind of example?

Well, basically you are feeding the algorithm with informations before you start the regression. The prior information is something like a base value the regressions starts with. For RAPM the previous season prior informed RAPM is used as the base level. Now, the algorithm takes the first snippet and tries to assign values to each player within a range centered around the prior (the range is determined by the penalizing factor lambda, which is found via crossvalidation, higher lambda = bigger penalty). In each step a new "prior" is calculated for the players. That can be described as pulling the value of each player closer to the prior. In order to change the value, the algorithm needs a lot of "confidence" to do so. That can be seen in the non-prior informed values, in which the default prior is used, which is just zero for all players. If there is not enough data to support a certain big change, the value will be left close to zero. You can picture each player with a rubber band to his chest, which is fixed to a certain object on the court. The object can be moved, if enough force is used. Now, the player tries to move away from the object, the stronger he moves, the further away does he get from that point, but no player can move constantly, he has to make stops, and at each stop the player gets pulled by the rubber band closer to the object again. That is somehow fitting picture, I guess.
That's why the non-prior informed version has all players closer to the mean than prior informed, and the prior informed will be closer to the previous season prior informed early in the season.

The out of sample test showed that prior informed RAPM gives a significant increase in predictive power.

Hope, that was somehow helpful ...

[lorak]

Yes, that was helpful, thanks a lot

[coldfish]

I haven't gone into great detail understanding RAPM. Since its output is effectively saying a lead balloon can fly, I'm not sure I want to waste the time. FWIW, I have an advanced degree in engineering so I can handle the math, but I have seen this countless times where someone makes an absurd number of assumptions to come up with some theory and then stands by it.

Guys like Rerisen and Alucryts have done a good job pointing out a number of the invalid assumptions, but I'll add one that I don't think I have seen.

As a fan who loves to watch the X's and O's, its been my experience that teams take about 2-4 weeks to game plan. That is to say, for about 2-4 weeks after a team makes a personnel or scheme change, teams don't adjust. That period often leads to better play than the long term average. Once teams "get the book" on a player or scheme, then they try to take away its or his strengths and things become more of a struggle. The stats being thrown around in this thread are assuming that team gameplanning has no impact on game results.

How does this relate to the Bulls? Well, when Rose plays, teams have a book on him. The traps, the pick and rolls, etc. They know how to guard him and the Bulls with him at the helm. When Rose went out, teams played their base defenses. I can guarantee this. I didn't see any special wrinkles put in place.

I guarantee you that if Rose was to miss the rest of the season with a knee injury, teams would eventually come around to treating Deng like a first option and it would crush him. His dribble is still too high and court awareness too low. All of the adjusted numbers would go to crap. It would even hurt the defense because the increased number of turnovers and fast break points.

One other note, a lot of these stats are dependent on who replaces a player. With Rose, he is replaced by Watson, who is more than a competent defender. With Deng, at approximately 9 minutes in the 2nd and fourth quarter, he is replaced by Korver, the worst defender on the team. Thibodeau does this rotation every single game. If Thibodeau were to change it up and have Korver on the floor with Deng and Brewer on the floor with Rose, Rose would "look" like a better defender despite not playing better defense.

.....

Just in case you guys care, here is the Bulls rotation:
Start Rose / Rip / Deng / Boozer / Noah
Brewer and Gibson come in for Rip and Boozer with around 4 minutes in the 1st.
Asik and Watson come in for Rose and Noah to start the 2nd.
Korver come in for Deng at 9 min in the 2nd
Rose come in for Watson at 8min in the 2nd
Boozer and Noah come back in around 6min in the 2nd
Deng comes back in for Brewer with about 5-6min in the 2nd

Thibodeau tries to keep this same rotation in the second half, but will play hot players in the 4th when its obvious.

Its extremely rigid and it really screws with a lot of these stats.

[mysticbb]

I haven't gone into great detail understanding RAPM. Since its output is effectively saying a lead balloon can fly, I'm not sure I want to waste the time. FWIW, I have an advanced degree in engineering so I can handle the math, but I have seen this countless times where someone makes an absurd number of assumptions to come up with some theory and then stands by it.

Well, if you could handle the math, you wouldn't need to spend much time in order to understand RAPM. But I'm pretty sure that you can't even describe a Hilbert space exactly without looking it up on Wikipedia or so. ;) (I wouldn't even be surprised, if you don't have any clue at first what a "Hilbert space" could be.)

As a fan who loves to watch the X's and O's, its been my experience that teams take about 2-4 weeks to game plan. That is to say, for about 2-4 weeks after a team makes a personnel or scheme change, teams don't adjust. That period often leads to better play than the long term average. Once teams "get the book" on a player or scheme, then they try to take away its or his strengths and things become more of a struggle. The stats being thrown around in this thread are assuming that team gameplanning has no impact on game results.

Interesting. Do you have any evidence which supports your claim? Well, I bet that for each example which you can bring up in which a team plays better without an important player first and then becomes worse, I can bring you an example in which the team plays worse at the beginning and then become better. ;)

I guarantee you ... I can guarantee this.

I bet you can. :)

One other note, a lot of these stats are dependent on who replaces a player. With Rose, he is replaced by Watson, who is more than a competent defender. With Deng, at approximately 9 minutes in the 2nd and fourth quarter, he is replaced by Korver, the worst defender on the team. Thibodeau does this rotation every single game. If Thibodeau were to change it up and have Korver on the floor with Deng and Brewer on the floor with Rose, Rose would "look" like a better defender despite not playing better defense.

Well, if you would have checked out how many minutes those players are on the court together before you are making such a bold statement, you might as well wouldn't have brought that up. ;)

First of all, Korver is not the worst defender on the Bulls, that honor belongs rather to Boozer or maybe Rip Hamilton now, but Korver is for sure NOT the worst defender. Second of all: Rose played 33% of his playing time with Korver over the last 1.5 years, Deng played 30% with Korver. Rose played 38% of his time with Brewer, Deng played 37% of the time with Brewer. Well, you really believe that the Korver/Brewer thing is responsible for Rose "looking" worse defensively? Do you have an idea how much the difference would be, if we assume that Korver is the worst defender in the league while Brewer is the best?

Its extremely rigid and it really screws with a lot of these stats.

Well, you said you can handle the math, try to come up with an idea to compensate for this. ;)

[coldfish]

First of all, Korver is not the worst defender on the Bulls, that honor belongs rather to Boozer or maybe Rip Hamilton now, but Korver is for sure NOT the worst defender. Second of all: Rose played 33% of his playing time with Korver over the last 1.5 years, Deng played 30% with Korver. Rose played 38% of his time with Brewer, Deng played 37% of the time with Brewer. Well, you really believe that the Korver/Brewer thing is responsible for Rose "looking" worse defensively? Do you have an idea how much the difference would be, if we assume that Korver is the worst defender in the league while Brewer is the best?

I will just give one example of an assumption you are completely ignoring with the above paragraph. Go back and look at my rotation. Despite the minutes breakdowns, Rose plays the vast majority of his time against starters. He rests at the start of the second and fourth quarters when other teams have their benches in. Thibodeau uses Deng as the anchor for the second unit and then has him rest when the other starters are back in during the 2nd and 4th quarter.

I'm not going to give you a stat on it, because its patently obvious, but starters are better than bench players. All of your minutes break downs work on the ASSUMPTION that a large enough sample size will level out the quality of the opponents, but due to Thibs' "set in stone" rotation, that isn't true. To the best of my knowledge, no website tracks this though so statistical analysis is impossible. You just have to watch the games.

I know alucryts has synergy sports access and experience. I suspect that he could back up some of my claims in this thread by showing the number of times that Rose and Deng are double teamed, which would back up my "gameplanning" assertion.

Interesting. Do you have any evidence which supports your claim? Well, I bet that for each example which you can bring up in which a team plays better without an important player first and then becomes worse, I can bring you an example in which the team plays worse at the beginning and then become better.

If you are so confident in this, I will make you a bet. Jeremy Lin is currently sporting a 22.9 PER and the Knicks are 10-3 in his starts. I will make a sig bet that he nor the Knicks is able to maintain that pace for the rest of the season.

Its basically the same situation as the Bulls were in without Rose. A new set up against a bunch of weak teams yielding statistics which were not representative of true ability and performance.

[mysticbb]

All of your minutes break downs work on the ASSUMPTION that a large enough sample size will level out the quality of the opponents, but due to Thibs' "set in stone" rotation, that isn't true. To the best of my knowledge, no website tracks this though so statistical analysis is impossible. You just have to watch the games.

And that's where ridge regression comes in handy. If you would be as smart and knowledgeable about math as you think, you would even suggest using it in order to adjust for different strength of opponents. ;)

If you are so confident in this, I will make you a bet. Jeremy Lin is currently sporting a 22.9 PER and the Knicks are 10-3 in his starts. I will make a sig bet that he nor the Knicks is able to maintain that pace for the rest of the season.

First of all: What has that to do with your claim that teams are always playing good without a star player and then becoming worse over time, because the other teams adjust? I take it that you can't back up your claim here. In fact we saw multiple times teams breaking down with their best players out right from the beginning.

What kind of "bet" do you want to make? That I agree that the Knicks aren't likely to win 77% of their next 30 games? Well, I give you a hint, the numbers suggest that they will win less. But well, before I present the calculation, let me ask you how many wins do you expect.

Its basically the same situation as the Bulls were in without Rose. A new set up against a bunch of weak teams yielding statistics which were not representative of true ability and performance.

That's why we should adjust for strength of opponents and HCA, before we making any bold claims. ;)

[coldfish]

All of your minutes break downs work on the ASSUMPTION that a large enough sample size will level out the quality of the opponents, but due to Thibs' "set in stone" rotation, that isn't true. To the best of my knowledge, no website tracks this though so statistical analysis is impossible. You just have to watch the games.

And that's where ridge regression comes in handy. If you would be as smart and knowledgeable about math as you think, you would even suggest using it in order to adjust for different strength of opponents.

You can't do any analysis on in game opponent strength with the data available. I think that fact is escaping you.

First of all: What has that to do with your claim that teams are always playing good without a star player and then becoming worse over time, because the other teams adjust? I take it that you can't back up your claim here. In fact we saw multiple times teams breaking down with their best players out right from the beginning.

What kind of "bet" do you want to make? That I agree that the Knicks aren't likely to win 77% of their next 30 games? Well, I give you a hint, the numbers suggest that they will win less. But well, before I present the calculation, let me ask you how many wins do you expect.

A normal sig bet. If you lose, I pick your sig for a month and vice versa. I am confident that, given my knowledge of basketball and statistics, that neither Jeremy Lin nor the Knicks will continue their pace just like I was confident that the Bulls, without Rose, would have been able to maintain their win percentage nor point differential over the course of a season.

[mysticbb]

You can't do any analysis on in game opponent strength with the data available. I think that fact is escaping you.

What kind of data is not available? I have a file with all different lineup matchups for the whole season. Why should I not have the ability to use that in order to adjust for opponent's strength?

A normal sig bet. If you lose, I pick your sig for a month and vice versa.

That is the part I understand, but I have yet to see a number from you. ;)

Please, give me a number for the amount of wins you expect the Knicks to have over their last 30 games.

[coldfish]

What kind of data is not available? I have a file with all different lineup matchups for the whole season. Why should I not have the ability to use that in order to adjust for opponent's strength?

OK, how has the Deng bench group (Watson / Brewer / Deng / Gibson / Asik) done against opposing starters (all 5)? How many minutes total?

That is the part I understand, but I have yet to see a number from you.

Please, give me a number for the amount of wins you expect the Knicks to have over their last 30 games.

How many do I expect? Don't have a number but based on their Lin pace, it would be 23. I bet that at the end of the season, Lin has less than a 22.9 PER and the Knicks have less than 41 wins. Once teams adjust to him and competition improves, its going to hurt them.

[mysticbb]

OK, how has the Deng bench group (Watson / Brewer / Deng / Gibson / Asik) done against opposing starters (all 5)? How many minutes total?

I would need to rewrite a script in order to find out which players they played against were starters and which were not. But I can easily assume that they didn't play only minutes against a unit with all starters for the other team. Overall they played 213 minutes over the last 1.5 years and 4.5 points better than average after adjustment for the strength of the opponents (regardless of being a starter or not).
Why do you ask that question? Why should I reduce sample size?

How many do I expect? Don't have a number but based on their Lin pace, it would be 23. I bet that at the end of the season, Lin has less than a 22.9 PER and the Knicks have less than 41 wins. Once teams adjust to him and competition improves, its going to hurt them.

Wow, you are expecting less than 23 wins for them over the next 30 games? Seriously, that is your answer? Ok, in that case a bet makes no sense at all, because I expect them also winning less than 23 games over the next 30. With all your supposed experience and knowledge this is the best you can come up with? That is somehow a joke, given the fact that you attacked any kind of statistical analysis in the first place.

But well. Let me give you an example of a rather useful analysis and prediction based upon the performance level of the Knicks over the last 13 games (it shouldn't be too hard to follow given the fact that you claimed you can handle the math, thus I can assume you have decent skills in Calculus 101): The average strength of the opponents was -1.85, the adjustment based on HCA gives us 1.23 (Knicks played 9 at home and 4 on the road, HCA is 3.2). The average scoring margin was 5.69. Overall the Knicks played like a 2.61 SRS team. Such a team is expected to win 18 of their next 30 games against an average schedule. The Knicks are playing 0.62 SOS over those 30 games, that gives us an expected scoring margin of 1.99. That adjustment gives us now 17 wins as expectations, if the strength of the opponents follows a normal distribution. Well, the Knicks are a little less lucky and have to play a different distribution. Going through that game by game we get 16 wins and 14 losses for the Knicks.

And that all is based on the performance level of the Knicks over the last 13 games. Yeah, that performance level is not in line with an expectation of 23 wins. That is also a good example how deceiving the record can be. SRS plus HCA is a better predictor for the outcome of future games than record.

[coldfish]

OK, how has the Deng bench group (Watson / Brewer / Deng / Gibson / Asik) done against opposing starters (all 5)? How many minutes total?

I would need to rewrite a script in order to find out which players they played against were starters and which were not. But I can easily assume that they didn't play only minutes against a unit with all starters for the other team. Overall they played 213 minutes over the last 1.5 years and 4.5 points better than average after adjustment for the strength of the opponents (regardless of being a starter or not).
Why do you ask that question? Why should I reduce sample size?

OK, next question, how did you adjust for the strength of the opponent?