RealGM Top 100 List #11

[trex_8063]

People are getting a little silly in the bashing of WS/48, pretending it's not a per minute stat. e.g. Chris Anderson having slightly better WS/48 than Lebron---in playoffs only (lolsamplesize); not close in rs---doesn't seem terribly relevant when he's playing barely more than a third the minutes.

Clearly a stat with flaws (can you name a single stat that ISN'T flawed as far as its ability to accurately define and rank greatness?). And does anyone actually use a single stat to define and rank greatness? I hope not. They all have their little biases.....that's why we shouldn't use only one.

[acrossthecourt]

Official run-off vote: Garnett

Oscar or Garnett was tough for me, but now it's Kobe versus Garnett. I could link back to previous posts about Garnett, but I thought I'd share this:
Doing a weighted average of his games missed and the SRS change from ElGee for Garnett 2008 through 2013, you have a +4.5 player on a great team with stars even without him.

His average age is 33.1 years old for those stats. If you adjust for age so he's at his peak around age 27 or 28, I think that's close to +8. (Looked at a few aging curves for this but I don't know how well it applies to superstars.) That is indeed a great player.

So anyway, not saying KG had an equal supporting cast in Minny during the years specified; but it's not the large gap that it appears you're trying to imply with name recognition (look: Parker and Ginobili!). Both Parker and Ginobili weren't there until at least 2-3 years into the "early 00's", and weren't particularly noteworthy players until AFTER the "early 00's". And everyone else listed except for Bowen and Rose was either pre-prime or post-prime/twilight.

Post-prime, pre-peak, post-modern, proto-punk, etc. Robinson/Parker/Manu are still much better than the level of support Garnett had, and I don't think it's close. The support level only gets close when Cassell and Wally Z are healthy, but rarely happened at the same time. Bowen's a better version of Hassell, Szczerbiak is pretty limited but honestly young Ginoibli was at least as good (Szczerbiak had issues staying on the court, so it's not a simple minutes argument.) Then things get really dire. Relying on Ricky Davis on offense. Guys like Mark Blunt and Mark Madsen are significant rotation pieces. And when Cassell was healthy, they did go to the WCF.

I don't know for certain if I think "it's close", but take a season like '01 for example:

Post-injury 35-year-old David Robinson is the clear 2nd-best player (no one else even close): he's still a pretty excellent defensive anchor and rebounder, but in a limited-minutes capacity (<30 mpg at this point in his career); avg <15 ppg by this point, too.
3rd wheel is probably Derek Anderson: 2nd-leading scorer on the team w/ 15.5 ppg---also 4.4 rpg/3.7 apg---with decent shooting efficiency and respectable defense. To me a prime Derek Anderson compares fairly well to someone like current Wesley Matthews: totally decent player, though perhaps not someone you'd expect to be the third-best player on a contender (unless the top 2 are superstars or near-superstars).
Then:
Antonio Daniels
Sean Elliott (oft-ill or injured he'll miss 30 games, and is barely effectual when he does play here in his final season)
Malik Rose (misses 25 games)
37-year-old Terry Porter
35-year-old Avery Johnson
34-year-old Danny Ferry
35-year-old Steve Kerr

So yes, it's better supporting cast than '05-'06 Minny had, but it's at least close to being close. At the very least, the differences between the respective supporting casts is not something you'd expect to be responsible for 20-ish more wins and ~7 SRS points.

EDIT: Or if you DO think it's worth that much, it must largely be on DRob. And if you think this highly of 35-year-old DRob, then perhaps what Chuck Texas said about him is true and he should be getting more consideration around this point of the list.

Yes I do think Robinson should get consideration for the teen's and pretty soon. It's trickier with him because his scoring is actually worse versus good defenses (unlike Garnett), and there are those playoff problems and the short prime, but I do and I've said so previously.

An all-star caliber player like that is monumentally worth a lot more than a bunch of crappy players.

And we're comparing knuckleheads like Ricky Davis and Griffin to guys like old Porter and Kerr. Their role players are better even if they're ancient. But they do have a few guys in their prime like Antonio Daniels and Rose. I think most people gloss over the names pretty quickly and some stuff is lost over the years, but it's harder to separate these top 3 through 10 guys. Again, I go back to this a lot, but it's like Hickson versus Robin Lopez. Most people years from now wouldn't see a large difference between them, but look what happened to Portland.

If Duncan=Garnett
Robinson > Szczberiak+Hudson or something. It's not even close. Szczerbiak only played 1500 minutes.
Anderson > Ricky Davis (Davis is basketball cancer)
Daniels > Jaric (Jaric played terribly)
Malik Rose > Griffin (Rose was a pretty solid rotation big)
Porter > Marcus Banks (Porter was old, but I'd say he's at least a hair over Banks)
Ferry > Blount (old but Ferry was a stretch 4 and shot well, and Blount didn't really do anything well)
Avery Johnson+Elliot > Hassell
Kerr (650 minutes) > Anthony Carter (didn't play well)
Samaki Walker > Madsen


Besides, we voted in Duncan a long time ago, so if Garnett isn't as good as he is, he can still be 11th.

[Black Feet]

This thread is infested with Kobe "hate."

Most first team ALL-NBA selections of all time.

Most first team ALL-DEFENSE selections of all time.

5 rings as 1 of 2 most important players on the team.

Led the league in total points 4 different seasons.

31,700 regular season points and counting, which is the 4th most all time.

2 scoring titles and #2 in scoring 4 other times.

Top 5 in MVP voting in 11 straight seasons.

Led all 5 title teams in assists.

I apologize to anyone if I come off as offensive, but the disrespect toward Kobe Bryant in this thread is offending me. I'm going to stop visiting this thread before my blood pressure rises. When I see people claiming Kobe isn't top 10 all time or even mentioning Garnett in the same sentence with him, that's my cue. Once again, sorry for any disruption.

God bless.

Honestly looking at the list of voters I would be surprised if he even made top 15 lol. Don't let it bother you his peers recognize his greatness as well as coaches GMs and experts, also his accomplishments speak for themselves which is all that really matters at the end. His achievements blow away some in the top ten let alone the players that are left, It seems to get worse and worse everytime it's redone. I don't understand why people are obsessed with ranking players anyhow, people aren't obsessed with rankings in all the other sports.

[trex_8063]

If Duncan=Garnett
Robinson > Szczberiak+Hudson or something. It's not even close. Szczerbiak only played 1500 minutes.
Anderson > Ricky Davis (Davis is basketball cancer)
Daniels > Jaric (Jaric played terribly)
Malik Rose > Griffin (Rose was a pretty solid rotation big)
Porter > Marcus Banks (Porter was old, but I'd say he's at least a hair over Banks)
Ferry > Blount (old but Ferry was a stretch 4 and shot well, and Blount didn't really do anything well)
Avery Johnson+Elliot > Hassell
Kerr (650 minutes) > Anthony Carter (didn't play well)
Samaki Walker > Madsen

.

Yeah, OK, that seems about right.
I think '06 Griffin trumps '01 Malik Rose slightly, though. Rose wasn't yet into big rotational minutes in '01, and missed 25 games besides. Meanwhile Griffin, although atrociously bad on offense (I mean special kind of atrocious), was a shot-blocking machine and an excellent rebounder: 2.1 bpg in <20 minutes, 5.9 blocks per 100 possessions ; which going for that many blocks you'd think would often leave him out of the play for the rebound, yet he still had a very very respectable 24.3% DREB% (17.2% TREB%), avg 15.5 reb per 100 possessions (5.6 rpg in <20 minutes). Had DRtg of 99 (just 1 pt worse than KG himself).

Other than that comparison, I suppose I agree with the others (although with the exception of DRob>Wally + Troy, all of them are quite close).

How many wins is Gregg Popovich> Flip Saunders or Dwane Casey worth, I guess is what we should be asking.

[Notanoob]

This thread is infested with Kobe "hate."

Most first team ALL-NBA selections of all time.

Most first team ALL-DEFENSE selections of all time.

5 rings as 1 of 2 most important players on the team.

Led the league in total points 4 different seasons.

31,700 regular season points and counting, which is the 4th most all time.

2 scoring titles and #2 in scoring 4 other times.

Top 5 in MVP voting in 11 straight seasons.

Led all 5 title teams in assists.

I apologize to anyone if I come off as offensive, but the disrespect toward Kobe Bryant in this thread is offending me. I'm going to stop visiting this thread before my blood pressure rises. When I see people claiming Kobe isn't top 10 all time or even mentioning Garnett in the same sentence with him, that's my cue. Once again, sorry for any disruption.

God bless.

You don't sound like you came here with an open mind. I came to this website with a very different idea of who was better than who, but I read a lot and made an effort to say "hey, maybe I'm not giving Russel enough credit" and so on, and I felt like I learned something from it.

In any event, accolades, team success and volume scoring are not the top three things to base your evaluation of a player on.

Team success is dependent on management, not every great player catches the breaks to win a championship. What if Kobe isn't paired with loaded frontcourts and never wins a title because his GM was garbage? Does that make him less of a player? Of course not. So why is the opposite true? Basically, winning titles doesn't make you the better player.

Accolades are often driven by narratives and given to the wrong person. I don't think that this really needs an explanation.

Volume scoring needs context to matter, otherwise when are we voting in AI or Dantley? Moses scored more points in his career than Shaq, but nobody has him ahead because of that, and Moses had plenty of team success too. Heck, Dan Issel and Elvin Hayes (who?) both scored more career points than Dirk or Hakeem, I don't think they're even going to get ranked.

[Moonbeam]

A question about Prior-adjusted RAPM for those familiar with it: the name brings to mind a Bayesian estimation setting. Is that true, or is there simply some sort of multiplier for previous seasons included? If the estimation is indeed Bayesian, how are the priors determined for each player?

My understanding, others correct me if I'm wrong:

Yes, think Bayesian. That's certainly the initial influence. I can't speak to whether there's some aspect of what they're doing that's unorthodox on that front.

RAPM without a prior basically adds extra fake data to the real data (APM being just real data), with that fake data being like every player everywhere spent time with no scoring happening. This has a huge effect on any player whose rating is dependent on some very small sample set, and given the amount of luck in basketball and lineup rotations, this prevents us from going nuts over some trend that was actually just a fluke.

However, using 0s arbitrary. It's an obvious choice to use if you know nothing, but we don't know nothing. We have other data for these players, so why not use it.

My understanding is that anyone doing a "classic" prior-informed RAPM for a given season uses the previous seasons non-prior-informed RAPM instead of 0s. We if a guy had a +5 last year, that gets populated instead of a 0.

Thanks for the reply! I'm by no means an expert in Bayesian analysis, but I think it's not quite right to suggest that using uninformative prior distributions (with mean zero) is the same as using fake data - it simply means that there is no reliable information to serve as a starting point, whereas informing the prior through previous seasons is simply changing the starting point in the presence of better information. I agree that it's probably a good idea to used informative priors when sufficient data is available from previous seasons, but I still have a few questions:

1. What is done with the very first season for which the lineup data for RAPM is available? I imagine an uninformative prior is used.

2. It seems that when the previous season is available, the posterior mean is used as the mean of the prior distribution. Is this done for both offense and defense? And what is the actual distribution that is imposed - Normal? Something else? What about the other parameters (e.g. shape parameters impacting variance)? Are they the same for everyone?

3. What is done with rookies? Are they all given a prior mean of 0?

4. Is there some age effect included? I imagine putting some general effect of a typical career trajectory might help a little, though it may prove insignificant.

5. Is there some sort of burnin used before determining the posterior distribution? If so, what is it?

I'm sorry to ask so many questions - I'm genuinely curious about this, and with my little experience using Bayesian estimation, I find that the results can be quite dependent on the prior choice.

[acrossthecourt]

A question about Prior-adjusted RAPM for those familiar with it: the name brings to mind a Bayesian estimation setting. Is that true, or is there simply some sort of multiplier for previous seasons included? If the estimation is indeed Bayesian, how are the priors determined for each player?

My understanding, others correct me if I'm wrong:

Yes, think Bayesian. That's certainly the initial influence. I can't speak to whether there's some aspect of what they're doing that's unorthodox on that front.

RAPM without a prior basically adds extra fake data to the real data (APM being just real data), with that fake data being like every player everywhere spent time with no scoring happening. This has a huge effect on any player whose rating is dependent on some very small sample set, and given the amount of luck in basketball and lineup rotations, this prevents us from going nuts over some trend that was actually just a fluke.

However, using 0s arbitrary. It's an obvious choice to use if you know nothing, but we don't know nothing. We have other data for these players, so why not use it.

My understanding is that anyone doing a "classic" prior-informed RAPM for a given season uses the previous seasons non-prior-informed RAPM instead of 0s. We if a guy had a +5 last year, that gets populated instead of a 0.

Thanks for the reply! I'm by no means an expert in Bayesian analysis, but I think it's not quite right to suggest that using uninformative prior distributions (with mean zero) is the same as using fake data - it simply means that there is no reliable information to serve as a starting point, whereas informing the prior through previous seasons is simply changing the starting point in the presence of better information. I agree that it's probably a good idea to used informative priors when sufficient data is available from previous seasons, but I still have a few questions:

1. What is done with the very first season for which the lineup data for RAPM is available? I imagine an uninformative prior is used.

2. It seems that when the previous season is available, the posterior mean is used as the mean of the prior distribution. Is this done for both offense and defense? And what is the actual distribution that is imposed - Normal? Something else? What about the other parameters (e.g. shape parameters impacting variance)? Are they the same for everyone?

3. What is done with rookies? Are they all given a prior mean of 0?

4. Is there some age effect included? I imagine putting some general effect of a typical career trajectory might help a little, though it may prove insignificant.

5. Is there some sort of burnin used before determining the posterior distribution? If so, what is it?

I'm sorry to ask so many questions - I'm genuinely curious about this, and with my little experience using Bayesian estimation, I find that the results can be quite dependent on the prior choice.

Yes an uninformed prior. However, what's popular now is to use a prior that's a mix of the previous season's values plus a metric known as a statistical plus/minus built off a long-term RAPM study. Basically, it's mostly box score stats plus a few other things (some use height, and some are getting advanced with non box score stuff like screens and contested shots) that correlates to a previous set of RAPM values. Rookies are generally -2 (offense plus defense). It's one of the biggest issues. Some want to use draft position, but I think that's adding an incorrect systematic error: one use of RAPM is to find underrated guys, and we don't want to overrate lottery rookies. I've used a different penalty factor for rookies. Some split rookies by pure rookies from college/high school and rookies with pro experience (those guys are like -1.) Age effects are usually included in models now. It's done for offense and defense, but for non-public general discussion things aren't split into offense/defense. It's easier to just do overall +/- without splitting. I don't think anyone imposes anything else or uses burnin.

Here's one of the more advanced models out there:
http://talkingpracticeblog.com/

Machine learning, random forest, gradient boosting, etc. I don't entirely understand their process, however, although I have run some stuff with that.

[RayBan-Sematra]

Random question about Barkley: how do you view his play with other ball-dominant players? How was his pick and roll game?

I think it is probably pretty good.
I never got to watch much Prime Barkley but for a volume scoring offensive anchor he actually had a reasonably low usage rate throughout his Prime.
Some people say he was a ball stopper who would over dribble but from what limited footage I have seen of him that doesn't appear to be the case.
His team ORTG's were usually very good especially once he got to Phoenix where he was surrounded by the most offensive talent.
That doesn't suggest to me a guy who has a problem fitting in with other good offensive players.
To be fair though I don't remember Barkley playing next to any really ball dominant players.
Even a guy like Kevin Johnson usually had a usage rate below 25%.

Not sure about his effectiveness in the PnR. I would assume it was pretty good given his skillset.
ThaRegular did an offensive analysis of Barkley over some sample of games which would show you how good he was in the PnR (over those games anyway).
So you might wanna look that up. Too lazy to do it right now.

Most first team ALL-NBA selections of all time.

Era and position based accolade which he clearly didn't deserve in certain years.
Beyond that his longevity is not unusually good compared to earlier and current candidates.

Most first team ALL-DEFENSE selections of all time.

Subjective award which he clearly didn't deserve in certain years.

5 rings as 1 of 2 most important players on the team.

Team accomplishment achieved largely through having good luck and circumstances.

31,700 regular season points and counting, which is the 4th most all time.

Karl Malone has a few thousand more points but he isn't in our Top 10.

[BallerTed]

This thread is infested with Kobe "hate."

Most first team ALL-NBA selections of all time.

Most first team ALL-DEFENSE selections of all time.

5 rings as 1 of 2 most important players on the team.

Led the league in total points 4 different seasons.

31,700 regular season points and counting, which is the 4th most all time.

2 scoring titles and #2 in scoring 4 other times.

Top 5 in MVP voting in 11 straight seasons.

Led all 5 title teams in assists.

I apologize to anyone if I come off as offensive, but the disrespect toward Kobe Bryant in this thread is offending me. I'm going to stop visiting this thread before my blood pressure rises. When I see people claiming Kobe isn't top 10 all time or even mentioning Garnett in the same sentence with him, that's my cue. Once again, sorry for any disruption.

God bless.

Don't be too offended, these ranking are only a few peoples opinions. There never was and never will be a consensus all-time list. They could do the same project a month from now with different voters and the Top 10 would be different. Since the last project in 2011, only two players in the Top 10 kept their same spot with Jordan at #1 and Hakeem at #9. Magic and Bird both dropped 4 spots. I see these projects as fun discussions for fans of the game to debate and discuss, but in the end it's all subjective so don't take it too seriously.

God Bless

[ElGee]

ElGee, I'm looking at your WOWY data for Kobe, and I was wondering if you had the data for 04 and 10, controlling for Gasol in 10, and Payton/Malone/Shaq in 04...or did it become too small of a sample to look at?

In 2004, Kobe misses 3 games with everyone else in the lineup. He misses 7 games with Malone and Shaq and another 7 games with Malone out. The best we could do would be to isolate the 7 games without Malone and compare them to the 15 games Malone missed with the other guys playing: -2.1 without, +5.0 with. But enormous error in both those sample sizes (the in and the out).

In 2010, Bryant misses 3 controlling for Bynum and co. If you include the games that Bynum also missed, save the throw-away 82nd game (for 8 games), LA was +8.8 in those games (-9.7 Def). So if you're scoring at home -- and warning, this will give some people some angst and might bother them immensely:

-the 2010 Lakers when Gasol was out played like a 42-win team. (95% chance they were not better than 57-wins)
-the 2010 Lakers when Bynum was out played like a 47-win team. (95% chance they were not better than 63-wins)
-the 2010 Lakers when Bryant went out (even without Bynum) played like a 64-win team. (95% chance they were not worse than 44-wins.)

I'm also going to post some performances against good/bad defenses in the RS and PS on the PC board so people can reference it.

[john248]

What the hell happened in this thread.

People still have issues with RAPM. But the reality is that people have issues with stats in general and how to interpret them. RAPM, other +/-, TS%, oRTG, WS are all things that need narratives. Measuring defensive contributions will continue to be a point of argument. There's no set criteria either.

[Senior]

Era and position based accolade which he clearly didn't deserve in certain years.
Beyond that his longevity is not unusually good compared to earlier and current candidates.

What years do you not think he deserved it? He made All-NBA 1st in 02-04, 06-13. I honestly think he was one of the two best guards in 2001 (AI and Kidd made 1st that season but the Lakers had a bleh RS, then assassinated everyone in the playoffs). I mean...maybe he shouldn't have made 12-13, but who would you have over him? His competition those years was Paul who was the other guard on 1st team, Parker, Westbrook, Harden, Wade, Curry...and if you discount Kobe for defense you have to throw out Wade, Harden and Curry. Westbrook is really the only arguable one. 2011? Wade still had it that year, but you could argue 2 of Kobe/Rose/Wade. There aren't any years where he clearly didn't deserve it. His competition over the years has been Iverson, T-Mac, Nash, Paul, D-Will, Wade, Kidd, Ray Allen...that's some nice competition.

Re: longevity. He's been near his prime or at his prime from 00-13. Down years in 04 and 05, fell off defensively from 11-13 but his offense has always been there.

Subjective award which he clearly didn't deserve in certain years. (all D)

This one I agree a little more with, but you have to admit that the coaches and media didn't just randomly pick a player to give All-D teams, he obviously deserved it some years. 00-04 he was very good to elite. As his offensive responsibilities rose his defense fluctuated but he could still defend 1-3 when the Lakers needed him to (think 05-07). 08-10 since he didn't need to do everything on O he could commit to playing well on D again. His help D in the championship years was great, he could still lock people up in series (for example Westbrook in 2010) 11-13 he kind of sucked on help defense but could still play good man to man.

Team accomplishment achieved largely through having good luck and circumstances.

I don't get this team accomplishment thing, like Kobe wasn't a huge part of his championship teams. No one does it alone. It's not like he was Adam Morrison or even Robert Horry, this is a guy who averaged like 29/7/6 as a second guy to Shaq. Not exactly easily replaceable. I better not see you say something like "Shaq carried the 00 Lakers" by the way

You don't think Shaq was lucky to get a coach basically tailor made for him and three elite guards? You don't think Lebron got lucky that Chicago fell apart and Boston got old and everyone else except Indiana sucked? He made the most of his opportunities, that's all you can ask. All the greatest winners get lucky, but they also cash in on their chances.

Karl Malone has a few thousand more points but he isn't in our Top 10.

He's getting discussion this early. Wouldn't surprise me to see him at something like 16th.

[HeartBreakKid]

This thread is infested with Kobe "hate."

Most first team ALL-NBA selections of all time.

Most first team ALL-DEFENSE selections of all time.

5 rings as 1 of 2 most important players on the team.

Led the league in total points 4 different seasons.

31,700 regular season points and counting, which is the 4th most all time.

2 scoring titles and #2 in scoring 4 other times.

Top 5 in MVP voting in 11 straight seasons.

Led all 5 title teams in assists.

I apologize to anyone if I come off as offensive, but the disrespect toward Kobe Bryant in this thread is offending me. I'm going to stop visiting this thread before my blood pressure rises. When I see people claiming Kobe isn't top 10 all time or even mentioning Garnett in the same sentence with him, that's my cue. Once again, sorry for any disruption.

God bless.

Honestly looking at the list of voters I would be surprised if he even made top 15 lol. Don't let it bother you his peers recognize his greatness as well as coaches GMs and experts, also his accomplishments speak for themselves which is all that really matters at the end. His achievements blow away some in the top ten let alone the players that are left, It seems to get worse and worse everytime it's redone. I don't understand why people are obsessed with ranking players anyhow, people aren't obsessed with rankings in all the other sports.

Basketball is probably one of the least obsessed sports when it comes to ranking. As someone who used to be a prize fighter this statement is really ironic.

[Moonbeam]

My understanding, others correct me if I'm wrong:

Yes, think Bayesian. That's certainly the initial influence. I can't speak to whether there's some aspect of what they're doing that's unorthodox on that front.

RAPM without a prior basically adds extra fake data to the real data (APM being just real data), with that fake data being like every player everywhere spent time with no scoring happening. This has a huge effect on any player whose rating is dependent on some very small sample set, and given the amount of luck in basketball and lineup rotations, this prevents us from going nuts over some trend that was actually just a fluke.

However, using 0s arbitrary. It's an obvious choice to use if you know nothing, but we don't know nothing. We have other data for these players, so why not use it.

My understanding is that anyone doing a "classic" prior-informed RAPM for a given season uses the previous seasons non-prior-informed RAPM instead of 0s. We if a guy had a +5 last year, that gets populated instead of a 0.

Thanks for the reply! I'm by no means an expert in Bayesian analysis, but I think it's not quite right to suggest that using uninformative prior distributions (with mean zero) is the same as using fake data - it simply means that there is no reliable information to serve as a starting point, whereas informing the prior through previous seasons is simply changing the starting point in the presence of better information. I agree that it's probably a good idea to used informative priors when sufficient data is available from previous seasons, but I still have a few questions:

1. What is done with the very first season for which the lineup data for RAPM is available? I imagine an uninformative prior is used.

2. It seems that when the previous season is available, the posterior mean is used as the mean of the prior distribution. Is this done for both offense and defense? And what is the actual distribution that is imposed - Normal? Something else? What about the other parameters (e.g. shape parameters impacting variance)? Are they the same for everyone?

3. What is done with rookies? Are they all given a prior mean of 0?

4. Is there some age effect included? I imagine putting some general effect of a typical career trajectory might help a little, though it may prove insignificant.

5. Is there some sort of burnin used before determining the posterior distribution? If so, what is it?

I'm sorry to ask so many questions - I'm genuinely curious about this, and with my little experience using Bayesian estimation, I find that the results can be quite dependent on the prior choice.

Yes an uninformed prior. However, what's popular now is to use a prior that's a mix of the previous season's values plus a metric known as a statistical plus/minus built off a long-term RAPM study. Basically, it's mostly box score stats plus a few other things (some use height, and some are getting advanced with non box score stuff like screens and contested shots) that correlates to a previous set of RAPM values. Rookies are generally -2 (offense plus defense). It's one of the biggest issues. Some want to use draft position, but I think that's adding an incorrect systematic error: one use of RAPM is to find underrated guys, and we don't want to overrate lottery rookies. I've used a different penalty factor for rookies. Some split rookies by pure rookies from college/high school and rookies with pro experience (those guys are like -1.) Age effects are usually included in models now. It's done for offense and defense, but for non-public general discussion things aren't split into offense/defense. It's easier to just do overall +/- without splitting. I don't think anyone imposes anything else or uses burnin.

Here's one of the more advanced models out there:
http://talkingpracticeblog.com/

Machine learning, random forest, gradient boosting, etc. I don't entirely understand their process, however, although I have run some stuff with that.

I imagine rookies are hard to consider. I'd imagine players who miss an entire season due to injury are also difficult to place.

Thanks for the link to the random forest/gradient boosting method. I'm not quite sure how they don't include information from prior seasons but have stuff like 3 point rate in there, but it's still interesting to see ensemble methods thrown in there.

As for the prior distributions, do you know what the distributional form is, and what the relevant shape parameters are? That would be interesting to me as well. I imagine it's easier to get a gauge on players with more minutes played than others, hence I'd imagine the variance in their prior distribution should be smaller.

The lack of burnin time seems a little strange to me, particularly if none is used for the first season of available data, but if the Markov Chain Monte Carlo is run long enough (I imagine this stuff is done with MCMC?), I suppose it won't matter too much.

[Black Feet]

This thread is infested with Kobe "hate."

Most first team ALL-NBA selections of all time.

Most first team ALL-DEFENSE selections of all time.

5 rings as 1 of 2 most important players on the team.

Led the league in total points 4 different seasons.

31,700 regular season points and counting, which is the 4th most all time.

2 scoring titles and #2 in scoring 4 other times.

Top 5 in MVP voting in 11 straight seasons.

Led all 5 title teams in assists.

I apologize to anyone if I come off as offensive, but the disrespect toward Kobe Bryant in this thread is offending me. I'm going to stop visiting this thread before my blood pressure rises. When I see people claiming Kobe isn't top 10 all time or even mentioning Garnett in the same sentence with him, that's my cue. Once again, sorry for any disruption.

God bless.

Honestly looking at the list of voters I would be surprised if he even made top 15 lol. Don't let it bother you his peers recognize his greatness as well as coaches GMs and experts, also his accomplishments speak for themselves which is all that really matters at the end. His achievements blow away some in the top ten let alone the players that are left, It seems to get worse and worse everytime it's redone. I don't understand why people are obsessed with ranking players anyhow, people aren't obsessed with rankings in all the other sports.

Basketball is probably one of the least obsessed sports when it comes to ranking. As someone who used to be a prize fighter this statement is really ironic.

Well I'm talking about team sports where different positions exist. Individual sports like boxing makes more sense when ranking even then they usually rank them according to weight class. Ranking basketball players it would make more sense to break it down by position.

[Moonbeam]

How long do we have to vote in the runoff? I'm quite torn on the vote. I'm leaning slightly to Kobe despite far preferring KG as a player, but I'm quite unsure still.

[drza]

Spoiler:

The biggest issue is how RAPM is being used. I've tried to stay away from it but when its used to say pre-prime 98/99 KG was playing at a superstar level and higher than 06/07 Kobe, it just boggles the mind.

98-99 Garnett: 19.3 ppg, 4.3 apg, 9.9 rpg on 51.3% TS
06-07 Kobe: 33.5 ppg, 4.9 apg, 5.5 rpg on 56.8% TS

If people want tor reference RAPM, fine. But to pump up KG's longevity by using RAPM from the 90's and the 11-13 time, as if he was playing at a superstar level....just leaves me a bit blank for words.

It gets to the point that we can't argue because production, peer-review is off the table. Playoff performances...nope, can't use them either. What's left

Yeah but timeout -- are you saying that a 19-10-3 +0% TS player can't be better than a 34-6-5 +3% TS player?

On its face, that would be extraordinarily unusual. But i did look further, specifically to things like peer review that regarded 06/07 Kobe as a Top 5 MVP/All-NBA 1st player. KG wasn't regarded as elite in 98 or 99.

If someone wants to explain how 98/99 KG was superstar level or better than 06/07 Kobe, I'm all ears.

Let's take a crack at it.

First, I'd start with re-iterating that RAPM doesn't say that KG was better in 98-99 than Kobe was in 06-07. What it says is that Garnett's presence on the court correlated with a larger increase in the scoring margins for the Wolves than Kobe's presence did for the Lakers in those years. So next up, let's interpret.

Here are the box score numbers that you cited:

98-99 Garnett: 19.3 ppg, 4.3 apg, 9.9 rpg on 51.3% TS
06-07 Kobe: 33.5 ppg, 4.9 apg, 5.5 rpg on 56.8% TS

So clearly, Kobe is operating at a much higher volume and efficiency on offense. On first blush, you might think that Kobe was having a huge offensive impact while Garnett's should be lesser. If we check out the offensive RAPM numbers for those years, we see this (per Doc MJ's normalized spreadsheet):

Offensive RAPM
Kobe '06: +7.9
Kobe '07: +7.8
KG '98: +3.9
KG '99: +3.1

OK, but what does that scale mean? Well, if we look at the best 5-year peaks of all players between 1998 - 2012, only Steve Nash (+9.1) and LeBron (+8.1) had higher peaks than +7.9. Wade, Shaq and Kobe are the next 3 highest 5-year peaks on the list, and Wade and Shaq only had a higher ORAPM than +7.9 in one season each. In other words, Kobe's +7.9 and +7.8 are really friggin high ORAPM marks, pointing out that he was having a mega offensive impact those years (as we'd expect).

Garnett's combined ~ +3.5 in those two years is in-line with a much more modest, but still positive, group. Looking at the 5-year peaks, +3.5 would be right in the middle of Andrei Kirilenko (5-year peak +3.6) and Josh Howard (5-year peak +3.4). KG's own offensive 5-year peak (as measured by ORAPM) was +5.4. On the surface, then, a +3.5 value seems like a reasonable neighborhood for a 20 ppg/4.3 apg big forward. But we should look closer, to see if we believe that KG's correlation to +3.5 of offensive scoring margin increase is causative or not.

The 1998 Wolves built their offense around KG, Gugliotta and Marbury and finished the year with the #7 O-rating in the NBA. But, Googs missed half the year and never played another game in Minnesota afterwards. Thus, for the majority of the year it was Garnett and Marbury driving the ship. KG was the leading scorer of the duo (barely), while Marbury led the team in assists. This arrangement was replicated the following year, but with KG taking on a larger scoring role and Marbury getting traded halfway through the season for Terrell Brandon (The Wolves finished '99 with the 17th O-rating in the NBA). In both instances, KG played a lot of 2-man games with the PGs as he thrived in the pick-and-roll/pop game. KG was already playing an attenuated version of the frontcourt offensive hub role, with a solid amount of distribution responsibilities.

In short, I'd think that a +3.5 ORAPM value would make sense for a player contributing his stats with his integral role in two offenses, one strong (top-7) and the other about average (#17).

Mini conclusion: the offensive RAPM scores for both Kobe and KG make sense given their production levels and levels of responsibility within their offenses. I believe that in both instances, the correlations were indeed causative. That Kobe really was having some of the best offensive seasons of the past 15 years, while KG was making a positive but much more modest offensive contribution. So, let's look at the other side of the ball.

Defensive RAPM
KG '98: +5.6
KG '99: +4.9
Kobe '07: -0.65
Kobe '06: -1.61

Alright, same drill. What do these numbers suggest about defensive impact, and do we buy them.

Kobe's defensive RAPM for those two seasons are negative. Does that make sense? These are the famous Kobe 1-man offense teams, where he clearly was focused on offense. Anecdotally, these are the seasons where many observed him coasting on defense to have more energy on offense. The Lakers were 15th and 24th in defense those two years, but I don't get the impression that defense was his concern in those two years. As a matter of scale, a composite defensive RAPM of about -1.1 in those two years would a small net negative but not a huge difference maker. I buy it.

KG's normalized defensive RAPM for those two seasons would be about +5.3. This is in the neighborhood of the 5 year peaks of Ben Wallace (+5.5), Bo Outlaw (+5.3) and Andrew Bogut (+4.9). Shane Battier is the highest rated defensive small forward from '98 - '12 with a 5-year peak of +4.3 and a single-season best of +5.6.

So, do we buy that late 90s KG could be having a defensive impact on the area of the best of Battier, on the continuum between the 5-year peaks of Battier and Big Ben (not including 2001 and some info from 2002). Is that a reasonable defensive neightborhood for that version of KG? Well, here is what I said about KG's 1-on-1 defense of wings and bigs in that time window, as well as what I said about his team defense at that portion of his career:

Early 1-on-1 wing defense:

Spoiler:

Early Garnett was the most explosive and athletic, but also the lankiest. Late 90s thru early 2000s KG played a lot of small forward. He was the full-time cover for several elite wings, usually to good result (though I recall Jordan torching him in his Wizards year...of course, even Wizards Jordan was quick for a SF and by then KG really should have been at full-time PF). He was maybe a step slower in terms of lateral movement than the best of the small forwards, but he was still ridiculously quick for a true 7-footer and he used his mantis arms and angles to excellent results. He could play a bit further off to discourage the drive, while still getting a threatening hand in the face of s ahooter. He could funnell drivers to where he wanted them to go, leading them into traps or difficult shots on the move. The most famous example of KG's defensive efforts on a wing were on Tracy McGrady in the 2003/2004 time period.

Over those two years, if you recall, McGrady was peaking...he led the NBA in scoring both seasons (35.4 ppg on 56.4% TS in 2003, 28 ppg on 52.6% TS in 2004). The Timberwolves played the Magic four times in that two-year window. In 2003 Garnett was the primary defender on McGrady, whereas in 2004 he was more of the main help defender. In all four years the Timberwolves devoted their main team defensive efforts to stopping McGrady (as pretty much all teams did at the time), so I don't want to give the impression that KG was just out there on an island with McGrady. Nevertheless, KG played the lead role in dramatically reducing TMac's output in those games. Of the four games, McGrady had two good games and two terrible games on his way to averages of 21.5 points on 49.6% TS . From the write-up to one of the terrible games:

Kevin Garnett recorded his seventh career triple-double and shut down Tracy McGrady. (snip)

Guarded mostly by Garnett, McGrady struggled with his shooting touch all night. McGrady, who scored 31 and 24 points in Orlando's first two games -- victories over Philadelphia and Miami -- went six for 15 for 18 points, 14 in the second half.

"That's the best I've ever been defended," McGrady said. "Ever.""

http://articles.latimes.com/2002/nov/02/sports/sp-nba2

Early prime 1-on-1 big man defense:

Spoiler:

By the time the 90s were coming to an end, KG was in his early 20s and starting to fill out. His listed weight went from the 220 pounds of his rookie year up to about 253 pounds by 2004. The 2003 season was the last year that KG spent a significant amount of time playing small forward, and by the fall of 2003 he was settling in as a full-time 4. In one-on-one circumstances, this version of KG was excellent on every big man south of Shaq. KG could really lock into post-scoring threats like Tim Duncan, versatile talents like Chris Webber, or even more perimeter based bigs like Rasheed Wallace. This versatility would serve him extremely well in this era, as the 2000s have been characterized by much more diversity at the 4 slot...from pure stretch 4s all the way down to more old-school post-up types. KG had the length and quickness to play great post-denial defense, making entry passes very difficult. He had to do his work early to prevent post position because he still wasn't the heaviest player, but even when post-players got position it was still very difficult to finish over those extendo-arms. And on the flip side, Garnett also had the quickness to hound his man all the way out to the 3-point line and beyond. He may have no longer been quick enough to guard small forwards full time, but he was still very possibly the quickest big man in the NBA. Good examples of KG's defense on the two extremes came in the 1999 (Tim Duncan) and 2000 (Rasheed Wallace) playoffs.

1999 Playoffs: Duncan averaged 15.9 pts/36 on 51.6% TS against KG, then 20.5 pts/36 on 58.8% TS against everyone else on way to title

2000 Playoffs: Wallace averaged 11.5 pts/36 on 57.1% TS against KG, then 19.2 pts/36 on 55.1% TS against everyone else

In Sheed's case the scoring efficiency was similar, but his volume was down by almost 40%. In Duncan's case, KG limited both his efficiency (7.2% TS difference) and volume (~23% down).

Young KG (Up through 2001) played during the time before the illegal defense rule change.

Spoiler:

He was playing a lot of both forward positions, so sometimes his man was on the perimeter and sometimes he was guarding more post-oriented players. But even when he was playing the best-of-the-best, his head was always on the swivel for help opportunities. The Wolves played a pretty vanilla defensive style, but KG gave them a disruptive wild card. This is when he was most athletic, and he used his long arms aggressively in the passing lanes and to block shots. Unlike Duncan, who seemed to get a lot of his blocks in 1-on-1 defense opportunities, Garnett's blocked shots seemed to come most often as weakside or topside rotations. It made for interesting angles on the Wolves defense, having their best shot-blocker often swooping down to the rim from the perimeter. This was when Garnett was setting his career highs in steals and combined steals and blocks.

I buy that a defender of this type could be having an impact similar to the best of Battier or high level Ben Wallace. But as UnbiasedFan pointed out, '98 and '99 wasn't getting those kinds of defensive accolades. While this is true, it should be noted that KG would be NBA All Defense 1st team in the next 6 consecutive seasons starting in 2000, AND that he'd finish second in the DPoY vote in both 2000 and 2001. He'd also finish as the MVP runner-up to Shaq in 2000, just a year away. So for those looking for the accolades to go with these big defensive marks, I think the argument that the voters just hadn't realized what KG was doing on defense (or overall) yet is reasonable.

Mini Conclusion: The only way that KG would be able to close the large difference in offensive impact is if he was having a huge defensive impact. His defensive skill set in those years was strong but unique. He was having the best combined block/steals numbers of what would eventually be a historic defensive career, he was exhibiting excellent 1-on-1 and team defense, and he was only a year away from being recognized as a top-2 defender in the NBA for those that appreciate accolades. I buy that his correlation with strong defensive results in '98 and '99 is causal, and therefore that RAPM did a good job of identifying defensive value that no other method of analysis would have been able to identify/quantify in the way that RAPM can.

So, does this mean that '98 and '99 KG was really BETTER than '06 and '07 Kobe?

I say no, it doesn't mean that.

"Goodness" and "better" are somewhat nebulous concepts. But while I believe that they should include "impact on team results" within the definition, that isn't the whole of the definition. And how big of a part it should play is an individual decision among analysts.

"Impact" as measured by something like RAPM or WOWY is going to be somewhat situation-dependent. We know from future years that KG's strong impact continued for the next decade-plus after 1999 in a large array of roles, but in '98 and '99 he wasn't yet what he would become. His jumper was solid, but it was a bit flatter and not nearly as consistent as it was by '03. His post moves and finishes were more raw. He wasn't the offensive hub/distributor yet that he would someday become. And IMO he hadn't developed that "I belong here" feeling that the truly great possess. In '97 the Wolves were overjoyed just to make the playoffs, and in '98 they had a sense of "nothing to lose/too young to be scared" when they pushed the 60-win Sonics to the brink in the first round of the playoffs. It really wasn't until KG faced off with Duncan in the '99 playoffs that I really got the sense that he was starting to come into his own on that level. And I think that's important.

Kobe, on the other hand, by 2006 knew that he was a superstar. He could score in every way known to man, and he expected to. He already had three rings on his hand, and it was his mission in life to prove that he could carry a team without Shaq which piqued his already obsessive competitiveness. The only reason that his RAPM values weren't higher in '06 and '07 was because of his defense, and 1) teams can and often do build strong defenses independently of their SG and 2) Kobe showed in the 3 years following (2008 - 10) that even if he wasn't a huge defensive positive, he could at least be a net neutral/small positive on defense while maintaining that mega offensive impact.

So in this case, it makes little sense to me to let the impact on team results make the whole decision. Yes, KG's impact on the scoring margins of the late 90s Wolves was seemingly on a similar order to the effect Kobe was having on the scoring margins of those mid'00s Lakers. However, that doesn't necessarily mean he was a better player. On the other hand, the fact that late 90s KG correlated to (slightly) more team success in his situation than mid-00s Kobe did to his also isn't an indictment on RAPM itself. When put into full context, the numbers fit reasonably into what played out on the court

[Baller2014]

How long do we have to vote in the runoff? I'm quite torn on the vote. I'm leaning slightly to Kobe despite far preferring KG as a player, but I'm quite unsure still.

If you're unsure, I'd encourage you to go back and read my Dr J v.s Kobe post on page 1. It's very troubling and telling that nobody has even tried to address the negatives Kobe brings with him.

I saw one poster bemoan that someone with Kobe's accolades, like most all NBA 1st teams of all time, has not been selected yet. Know who is tied with Kobe for most first teams? Karl Malone, and he hasn't gotten a single vote yet. Karl Malone's achievement is also more significant, because his position was generally tougher (when Karl Malone made a first team, there was no question he was a top 5-6 player in the NBA, whereas some years when Kobe made it he clearly wasn't). Karl Malone has 2 MVP's, not one, and basically beats Kobe everywhere. I can at least understand the anti-Dr J argument (though I don't agree with it), but what's the anti-Karl Malone argument? Karl is as good on offense as Kobe (especially when we factor in efficiency), vastly more impactful on D, and has more longevity. He seems to have a clearly stronger case than Kobe, yet he's got 0 votes. How is that?

[Moonbeam]

Excellent post, drza. Thanks for putting RAPM in context.

One other point that I feel deserves a mention, if I'm getting it right:

I imagine that the RAPM statistics cited are simply the means of the individual posterior distributions. In other words, a player with an RAPM of 7 may have a 95% credible interval from, say, 6 to 8, while someone with an RAPM of 6 may have a 95% credible interval from 4 to 8. In other words, the difference in posterior means may not be statistically significant, because of the standard errors associated with the estimation.

Can someone more familiar with RAPM confirm this?

[Moonbeam]

How long do we have to vote in the runoff? I'm quite torn on the vote. I'm leaning slightly to Kobe despite far preferring KG as a player, but I'm quite unsure still.

If you're unsure, I'd encourage you to go back and read my Dr J v.s Kobe post on page 1. It's very troubling and telling that nobody has even tried to address the negatives Kobe brings with him.

I saw one poster bemoan that someone with Kobe's accolades, like most all NBA 1st teams of all time has not been selected yet. Know who is tied with Kobe for most first teams? Karl Malone, and he hasn't gotten a single vote yet. Karl Malone's achievement is also more significant, because his position was generally tougher (when Karl Malone made a first team, there was no question he was a top 5-6 player in the NBA, whereas some years when Kobe made it he clearly wasn't). Karl Malone has 2 MVP's, not one, and basically beats Kobe everywhere. I can at least understand the anti-Dr J argument (though I don't agree with it), but what's the anti-Karl Malone argument? Karl is as good as offense as Kobe (especially when we factor in efficiency), vastly more impactful on D, and has more longevity. He seems to have a clearly stronger case than Kobe, yet he's got 0 votes. How is that?

Well, I'd argue that Karl is right in the discussion with both of them - I voted for Jerry West for this spot, so I'm considering 2 candidates I have further down on my personal list.

Also, Kobe is my least favorite player in NBA history, so I'm in full agreement with acknowledging the negatives that he brought on and off the court. I'm trying to keep my personal opinions of players' likeability out of the vote, and I have to somewhat agree with those who say that while Kobe's drawbacks did have an impact on his team's performance, his positive contributions as a player were still enough to be part of gaudy success. I fully believe that had he been more mature, Kobe could be sitting on 7 or more rings now... And as far as Kobe vs. Dr. J, I'm planning on voting for the Doctor over Kobe when the time comes.

The KG posts have been enlightening and have certainly made me reevaluate him - before this project I would have clearly had Kobe as the victor. I am a little hesitant to throw myself fully behind RAPM without getting a little more detail on the mechanics of it first, but it seems to be a worthwhile consideration that highlights contributions previously unheralded in box scores. I'll be reading through this thread again and weighing it up before casting my vote.

On a completely unrelated note, let me add that a little part of me dies every time someone brings up Adrian Dantley, my childhood hero and the reason I first became passionate about basketball, as a pariah of misleading value.