#25 Highest Peak of All Time (McGrady '03 wins)

[therealbig3]

Voting:

03 T-Mac - 4 (therealbig3, C-izMe, Dr Positivity, fatal9)

98 Malone - 2 (JordansBulls, Lightning25)

11 Dwight - 1 (DavidStern)

62 Baylor - 1 (PTB Fan)

[therealbig3]

So basically Malone's RAPM is based on incomplete data and not in his prime/peak years. In fact 3-4 after his peak. Why should it matter then is beyond me.

Well, he was still playing the same role, and his box score production barely fell off, and a big part of his all-time ranking is based on his longevity...which implies that he maintained a strong level of play for a very long time. So data 3-4 years after his peak would theoretically still give us an idea of how he played at his peak.

However, I think I agree that it's just one number, that's not even all that horrible (+2-3 a few years after his peak is not an unimpressive showing at all), and I think objectively looking at his overall game reveals an excellent player that deserves strong consideration at this level.


However, T-Mac has been criticized (imo, unfairly and without much evidence) that he struggled against top defenses...what about K. Malone? I'm not even saying it's a flaw, because I do understand how a unipolar offense with weak offensive support around the star can really drive your efficiency down...but the criticism levied against T-Mac is more apparent when you look at Malone, so looking at it objectively...K. Malone has the same issues that T-Mac does, at an even more extreme level. So the guys not voting for T-Mac...I don't really understand why you would vote for Malone. Even if he has a better defensive impact than a motivated T-Mac would...how much of a difference is there? Wade and Kobe have decent defensive impact when they turn up their defense, and guys with similar builds as T-Mac (LeBron/Pippen) have wreaked havoc defensively when they were dialed in. We've seen T-Mac play stretches of very strong defense as well, when he was able to. It's very hard for me to believe that K. Malone would be such an improvement defensively over T-Mac that the (imo) significant offensive gap can be overlooked.

Meh, food for thought.

[Doctor MJ]

So basically Malone's RAPM is based on incomplete data and not in his prime/peak years. In fact 3-4 after his peak. Why should it matter then is beyond me.

Not sure what to say here. I've repeated ad nauseum now that I agree that there's a problem with the data and that since my first post was talking about that data, it means the analysis it did based on that data needs to be taken back, so I kinda just want to leave this be but...

3-4 years after peak? Malone's 2nd MVP was in '99, we have the data from '01 when he had roughly the same box score production. You really don't think reasonable for people raise eyebrows when the data we have for that year doesn't come off looking that great?

[mysticbb]

He definitely used it in some fashion; he didn't just pull those records out of thin air.

Sorry, but Hollinger's team predictions are not based on individual PER, but on team offense and defensive efficiency analyses and predictions. PER might have an influence on the offensive efficiency (Hollinger does not disclose the used method fully), but Hollinger knows too well that his PER fails in terms of accounting for defense.

He might have also used other sources but that is unknown.

No, it is not unknown that he is using other sources. YOU don't know that! There is a difference between your lack of knowledge and the knowledge other people possess. If you had read his articles, you may have noticed that he is basing the prediction of the wins on the offensive and defensive efficiency prediction and part of the "other sources" is also his gut feeling about certain teams.

Others have also used their own metrics and managed to handle themselves well next to RAPM/box blends.

Which again doesn't say much, because not all used the same minutes distribution.

Of course I know that minute allocation is important here (which is why I stated "I don't use PER over RAPM because of the predictions").

Why haven't you mentioned the minute allocation even once before? And why are ignoring it in this very post again?

Except that's wasn't my point.

I stated a fact, nothing else. Given the amount of mistakes you are making, it was not unreasonable to assume that you included the "OLDRAPM" values in your argumentation against RAPM.

What's the difference? Half a win's worth? Is RAPM blowing everything else away, and allowing stats guys to make perfect predictions just using RAPM so they can cash in on their early retirement?

Given the fact that RAPM is adjusted for the strength of schedule, the value might even underestimate RAPM here, because Alex is comparing it with the MOV and not with SRS.
The 2nd part is again based on your lack of knowledge. Knowing minutes allocations, which includes making a good predictions on injuries, is part of that as well. Also, cashing in means beating Vegas, and Vegas is usually not that bad. And even if you can beat them on a regular basis, you will not that easily make a lot of money with that, because it is still also subject to variance.
You also seem to have a pretty tough time to understand that nobody is actually using RAPM as the lone evaluation tool, but rather as a part of the player evaluation.

...RAPM is one (albeit valuable) tool in the toolbox.

Alright.

Why did you repeat that? Did I disagree with that part? If I'm not responding to a part of your post, I might as well agree with it or think that this is actually a non-issue. In that case I agree.

[mysticbb]

So basically Malone's RAPM is based on incomplete data

Which data and RAPM values are you referring to?

[lorak]

3. In 1998 Jazz offense was nothing special without Stockton.

For whatever reason, they began playing better after the first 6 games when they made a switch in their starting SF (Russell to Keefe). They went from 2-4 to 9-3 after making this change (5 of those wins against playoff teams). That was their starting lineup for rest of the season with Stockton. In the 12 games that Malone played with that starting lineup, but without Stockton, Jazz appear to post a +4-5 offense (does anyone have the exact ortng for those games?), which would make them top 5 offense in the league.

Without Stockton Jazz had offense at 7th/8th place in the NBA. With Stockton they had offense BY FAR the best in the league and one of top 5 all time.

Also, why you ignore question about Malone's defense in comparison to McHale, Dwight and Hayes? Is Karl advantage on offense big enough to overcame defensive advantage each of these three players have over Malone?

And one more thing. I'm watching a lot of Jazz games recently and Karl was very good passer and overall team player on offensive end (just as you said in your long post), but his scoring in the playoffs was awful. I just watched G7 vs Sonics in 1996 and Malone was trying only tournaround fadeaway jumpers over and over again. He was also horrible on FT line (Sonics fans counting time ) and that are issues he showed through his whole career - not only during that game (finals vs Bulls or even games when he was hot, like the one I linked above to the youtube clip)

His defense also looked bad, or maybe it was Kemp who is underrated post player? Because he scored on Malone any time he wanted and was doing it often playing back to the basket. Perkins also had a lot of success when was guarded by Karl.

So overall I see prime Malone as average or slightly above average defensive player (I'm talking about whole impact on D, not only man to man D) and very good offensive team player, but whose scoring abilities were limited and often exposed in the playoffs. He simply liked his jump shot too much and that hurts overall value of player whose main strength is scoring.

That's why I don't see how he can be chosen over Dwight, who was worse offensive player, but much better defender or over McHale, who was comparable offensive player and also better defender than Malone. Karl for me is in a ballpark of Hayes - Malone was similar scorer, but better team player on offense, but also worse defender than Elvin.

And we also have several "small" players who probably should have been chosen over Karl, like Barry, King, TMac, Penny, Drexler or maybe even KJ and Pippen. (+difficult to evaluate players like Baylor, Gilmore and Pettit)

[bastillon]

therealbigthree, you need to stop with those Pippen/LeBron comparisons. Amare is somewhat built like Hakeem but it doesn't make him anything more than a horrible defensive player. TMac never showed great defensive team impact, whether his effort was there or not. during his 03 season he was a very bad defensive player according to lineup data. so when he goes all out offensively, his defense becomes an issue. that makes his offense kind of less valuable than it should be. I'm not buying TMac's impact at all. skillset is there, talent is there, just effort isn't. I could see how people can be fooled by his gametime because TMac's game is just so fluid and beautiful. but two-way impact just isn't there.

vote: 1988 McHale

[C-izMe]

therealbigthree, you need to stop with those Pippen/LeBron comparisons. Amare is somewhat built like Hakeem but it doesn't make him anything more than a horrible defensive player. TMac never showed great defensive team impact, whether his effort was there or not. during his 03 season he was a very bad defensive player according to lineup data. so when he goes all out offensively, his defense becomes an issue. that makes his offense kind of less valuable than it should be. I'm not buying TMac's impact at all. skillset is there, talent is there, just effort isn't. I could see how people can be fooled by his gametime because TMac's game is just so fluid and beautiful. but two-way impact just isn't there.

vote: 1988 McHale

I would argue that pre 03 TMac was a good defender. Not great but he was still a young wing playing for PT and in Orlando playing SF in a more balanced offense.

[MisterWestside]

PER might have an influence on the offensive efficiency (Hollinger does not disclose the used method fully), but Hollinger knows too well that his PER fails in terms of accounting for defense.

Which one is it mystic? Was it used or not?

No, it is not unknown that he is using other sources. YOU don't know that!

Doesn't seem you know much either, since you can't decide if PER was used or not. Heck, you know what -- fine, you're right. Holinger didn't use PER. Which means whatever "inferior" method/metric he was using, it still beat out the legendary RAPM.

Which again doesn't say much, because not all used the same minutes distribution.

Of course you'd say that. RAPM didn't win out.

I stated a fact, nothing else.

And the fact you stated was irrelevant to the point I made.

Given the fact that RAPM is adjusted for the strength of schedule, the value might even underestimate RAPM here, because Alex is comparing it with the MOV and not with SRS.

"Might". Are we going with conjecture now? Show this, Grand Poobah. Where's your retrodiction blog? Funny how someone else had to even do the work for you

And when you do this, show that RAPM blows everything else out of the water. Show it's the clear-cut, definitive metric.

Also, cashing in means beating Vegas, and Vegas is usually not that bad.

Nothing to do with Vegas here. And given your arrogance, I know if you came up with the all-in-one metric to perfectly model basketball and predict/analyze records and performance, you would have all 30 NBA teams eating out of your hand, not on these boards "correcting" people with your unerring knowledge.

Why did you repeat that? Did I disagree with that part?

Because you continue to come to its defense. RAPM can never be incorrect about a player. Nope

[mysticbb]

Heck, you know what -- fine, you're right.

That is the only sensible sentence you wrote in the whole post. And I don't think it makes any sense to discuss anything with you, because it seems as if your emotional attachment to the things you write is making it impossible to have a useful conversation. I'm not writing posts to massage your ego, if you can't handle it when someone points out that you are at fault, you might just look for a different place to express your opinion.
Btw, Engelmann (the guy providing us with the RAPM stuff) finished ahead of Hollinger in the prediction you are referring to. I didn't mentioned it before, because that prediction says really not much (if anything at all) about the quality of the models.


A disclaimer: RAPM is not flawless, I don't have the perfect model to describe a basketball game. Given the fact that a model is always just an approximation of the reality, I'm sure that something like that will never exists.

[MisterWestside]

That is the only sensible sentence you wrote in the whole post.

Of course, typical mystic Just take the part where I said "you're right" and then talk about egos? Oh, the ironing.

Btw, Engelmann (the guy providing us with the RAPM stuff) finished ahead of Hollinger in the prediction you are referring to

Here goes nothing. These were not done with RAPM, it's just me guessing

http://apbr.org/metrics/viewtopic.php?f ... 0&start=90

Yep, claim others don't get their facts right and mislead, then do the same thing yourself.

RAPM is not flawless

Good. Now move along.

[tsherkin]

Mister, mystic, play nice or don't talk to one another.

*waves finger, Deke-style*



Sent from my BlackBerry 9700 using Tapatalk

[mysticbb]

Good. Now move along.

My point still stands however. You're quick to explains flaws with data when it doesn't suit your argument. You didn't point out caveats with the use of any of the rother numbers.

And no, that doesn't mean that I think RAPM would be flawless, not at all. It is just simply the best available non-boxscore method we currently have.

Written on Wed Jul 25, 2012 10:25 pm.

viewtopic.php?f=64&t=1197924&p=32872531#p32873014

;)

Yep, claim others don't get their facts right and mislead, then do the same thing yourself.

Where did I claim that he did it with RAPM? ;)

[ardee]

Vote: 2003 T-Mac

There doesn't seem to be enough Malone backing, and anyway, I'm not the biggest fan of 1998.

I'm not getting too involved in the RAPM discussion, but from what I seem to have gleaned about the stat:

1. It doesn't take into account how the player's team performed when he was OFF the court.

2. It only takes into account the overall scoring margin when he is ON the court.

3. It uses certain statistical functions like regression to determine how much that particular player contributed to that scoring margin.

I just don't see it. To mathematically break down exactly how much a player was contributing while he was on the court is basically the same thing as working with other stats to produce numbers like PER and Wins Produced. It's from a different family of stats, but its still a stat that is derived through a variety of mathematical functions, just like PER.

Unless someone explained how exactly a mathematician calculates the extent to which a single player was responsible for a scoring margin, it can't be taken that seriously.

Something like on/off, for example, is a better stat, i.e. KG was +22.8 per 100 possessions in 2003. It's obviously not flawless itself, but its an accurate representation of what ACTUALLY happened, as compared to what someone THOUGHT happened, as in RAPM.

[mysticbb]

1. It doesn't take into account how the player's team performed when he was OFF the court.

As a part of the regression those minutes are taken into account. When you calculate such RAPM values, you always get a set of coefficients, one coefficient for each player, which are connected to each other. You can't calculate that without taking the minutes "off" into account.

3. It uses certain statistical functions like regression to determine how much that particular player contributed to that scoring margin.

The method is based on the regression analysis. R points to ridge regression.

And this method is far superior to the simple Net+/- you are referring to, because adjustments for teammates and opponents is included intrinsically, you can't remove that from the calculation.

[ardee]

1. It doesn't take into account how the player's team performed when he was OFF the court.

As a part of the regression those minutes are taken into account. When you calculate such RAPM values, you always get a set of coefficients, one coefficient for each player, which are connected to each other. You can't calculate that without taking the minutes "off" into account.

3. It uses certain statistical functions like regression to determine how much that particular player contributed to that scoring margin.

The method is based on the regression analysis. R points to ridge regression.

And this method is far superior to the simple Net+/- you are referring to, because adjustments for teammates and opponents is included intrinsically, you can't remove that from the calculation.

The "adjusted" form of +/- doesn't actually use "off court" at all. It's basically looking at what was accomplished by everyone when they were on the court, and distributing credit in the simplest way possible. There's more to a players abilities than this of course, but a player who consistently doesn't do great by this metric simply isn't showing signs that his team is benefiting immensely from his on court presence.

What is Doc trying to get across then?

[mysticbb]

What is Doc trying to get across then?

He just said that the adjusted version is unlike the Net+/- version. For Net+/- you subtract the OFF value from the ON value. When you run a regression and get APM-like values (RAPM is just a derivate here), you just get one number for each player.

BUT, you run the regression with all data included. When one player is off the court, someone else is on, obviously. Thus, all minutes are included in the calculation, also those minutes, in which a specific player was off.

[MisterWestside]

Vote: 2003 T-Mac

There doesn't seem to be enough Malone backing, and anyway, I'm not the biggest fan of 1998.

I thought that T-Mac should've been voted in from before anyway so good pick

As for RAPM, just know that you're posting with someone who can tell you about all the mathematical bells and whistles and how it "adjusts" for everything. But it's still flawed.

And as long as it (or any number) is flawed, use with caution.

[ElGee]

Ardee, read this (APM is at the bottom): http://www.d3coder.com/thecity/advanced-stats-primer/

The short of it is this: when two NBA lineups are on the court, there will be a margin of victory (MOV) from when they play. (One team "wins" or there is a tie.) For simplicity, you can think of the games as 2 on 2 games.

Team I (Player A, Player B) v Team II (Player C, Player D) results in a 7 point Team I win. The equation for that in the APM calculation is:

7 = (P1 + P2) - (P3 + P4)

We want to know the values for each player that are responsible for creating the 7 point difference. Was P1's value 10? 5? We don't know just from this information...so we look at another equation when Team I played Team III:

3 = (P1 + P2) - (P3 + P5)

In this game, the same team I beat a team II that swapped Player 4 for Player 5 by 3 points. Now we can make a guess at what the best values are for the 5 players to generate these results we see. eg

P1 +5
P2 +1
P3 +2
P4 -3
P5 +1

This gives us the scores perfectly -- there is no error in the actual MOV's from the data we plugged in with the values for the variables. The problem becomes when we add lots and lots of versions and lots and lots of lineups. For eg

-4 = (P1 + P4) - (P2 + P3)

The original captains swapped teammates and suddenly Team I lost by 4 points. The best player values from the first two games predicted a 1-point loss, but they lost by 4. Uh oh. Very quickly, we can no longer input a value where there is no error between our players and the results of their games...so we use (regression to attain) the "best fit" or smallest error when we plug in ANY player value to what happened in an actual lineup. In practicality, this is done w/ 5 on 5 lineups and HCA is accounted for as well as the minutes these lineups are on the court.

It would be a darn near perfect tool if we could increase the sample size and increase the number of lineups. But we can't so we are left with some instability in the numbers. Ridge Regression (RAPM) will reduce the error more than standard linear regression, but part of that comes at the expense of outliers; It's kind of "mashing" the numbers closer to 0. Single-season RAPM thus has some ballparking value -- everyone should be aware of the numbers -- but it's still subject to problems. Note: this is also why I wanted to know the errors (basketballvalue posts it's 2-yr APM errors...Jerry does not with his RAPM), because the Stockton/Malone case being raised here is really telescoping on that one 2001 result (Stockton +5.6, Malone +2.1).

If we considered the value per 100 pos (I believe this is what Jerry is posting) and incorporate number of pos played, the 2001 RAPM "Player Value Ranker" per game would look like:

Shaq 4.9
Duncan 4.8
Kobe 3.2
Dirk 3.1
Allen 3.1
Stockton 3.1
...
Bradley 2.4
Robinson 2.2
Outlaw 2.2
McKie 2.1
Fisher 2.0 (20 gp)
Anderson 2.0
McGrady 2.0
Brown 1.8
J. O"Neal 1.6
Miles 1.5
Malone 1.4

I left out 8 more notable players, so Malone would rank 25th (24th if we exclude Fisher), which isn't really a result that looks too strange. He may be "only" 2 years removed from his last MVP, but it's unanimously considered a down year after the lockout (the time off and compression of games affected older players), and in 2001 he was all-nba 3rd team...suggesting a ~top 15 player.

[Regulio]

So basically Malone's RAPM is based on incomplete data and not in his prime/peak years. In fact 3-4 after his peak. Why should it matter then is beyond me.

Not sure what to say here. I've repeated ad nauseum now that I agree that there's a problem with the data and that since my first post was talking about that data, it means the analysis it did based on that data needs to be taken back, so I kinda just want to leave this be but...

3-4 years after peak? Malone's 2nd MVP was in '99, we have the data from '01 when he had roughly the same box score production. You really don't think reasonable for people raise eyebrows when the data we have for that year doesn't come off looking that great?

Well we can look at Kobe Bryant, in 2009 he is a +6.1, in 2011 +2.7, in 2012 +1.5.
He regressed a bit, especially in 2012, but not really that much in boxscore stats.
How can we tell that Malone wasn't some +7 player in 1998, when Kobe went from +6.1 to +1.5 in 3 years ?
I don't think he was +7, but we don't have any evidence so why should we start guessing ?
That's my whole point