RealGM 2023 Top 100 Project - #8 (Shaquille O'Neal)

[Doctor MJ]

Induction Vote - 1

Shaq - 8 (OaD, beast, trelos, rk, tex, ceoofk, Narigo, cupcake)
KG - 8 (Dr P, hcl, iggy, ShaqA, eminence, DGold, speel, Ohayo)
Magic - 5 (Doc, f4p, AEnigma, falco, Moonbeam)
Curry - 1 (ltj)

No majority.

Induction Vote - 2

Shaq - 3 (ltj, f4p, AEnigma)
Curry - 1 (Doc)

This puts Shaq at 11-8 over KG. Note that that still doesn't give Shaq 1st or 2nd vote majority, though he does have half the total so even if we pressed the others (myself included) to choose, KG could only get it to a tie, so I don't think there's any point in debating finer details of voting. We could discuss further what we should do in a situation where, say, the leader only reaching 10 out of 22 even with 2nd votes. My intention was to just call it rather than delaying, but we could talk about it. For this time though, I'm calling it:

Shaquille O'Neal is Inducted at spot #8.

Nomination Vote

Kobe - 10 (trelos, ltj, rk, trex, ceoofk, ShaqA, f4p, AEnigma, speel, Ohayo)
Dirk - 2 (OaD, Dr P,)
Mikan - 4 (beast, eminence, cupcake, Moonbeam)
Oscar - 2 (Doc, Narigo)
Robinson - 1 (iggy)

Kobe Bryant is added to our Nominee list.

[Doctor MJ]

I've been playing around with box score data to run regressions on WOWY. Traditional WOWY statistics ignore what other players may be missing, so running a regression should offer an improvement. I've been looking at 5-year windows and have applied some shrinkage techniques like Ridge (as used in RAPM), Adaptive lasso and Adaptive Elastic net. I've been thinking about this with the view of voting for Magic here and thought I'd share the data I'm looking at before writing up my voting post. For simplicity now, I'll call this metric "RWOWY" for regressed WOWY, and then "RWOWY-Ridge", "RWOWY-Lasso", and "RWOWY-ENet" for the ridge, adaptive lasso, and adaptive elastic net variants.

These values make use of all games (regular season and playoffs) and all players who averaged at least 18 MPG for the season. I'll print out the results for 5-year spans over Magic's career for players who won at least one award (All-Star, All-NBA, All-Defense) during that span.

Immediate Hall of Fame RealGM post; you should make it into its own thread.

(And I appreciate having more material to use for my eventual Bob Lanier nomination. )

Are the excised rankings just players who played too few minutes to come across as notable? Might be appropriate to set a certain minimum games played threshold; I notice top rookies seem to do inordinately well.

Agree, I think a separate thread where people can discuss salient players in the data would be great.

[iggymcfrack]

Induction Vote - 1

Shaq - 8 (OaD, beast, trelos, rk, tex, ceoofk, Narigo, cupcake)
KG - 8 (Dr P, hcl, iggy, ShaqA, eminence, DGold, speel, Ohayo)
Magic - 5 (Doc, f4p, AEnigma, falco, Moonbeam)
Curry - 1 (ltj)

No majority.

Induction Vote - 2

Shaq - 3 (ltj, f4p, AEnigma)
Curry - 1 (Doc)

This puts Shaq at 11-8 over KG. Note that that still doesn't give Shaq 1st or 2nd vote majority, though he does have half the total so even if we pressed the others (myself included) to choose, KG could only get it to a tie, so I don't think there's any point in debating finer details of voting. We could discuss further what we should do in a situation where, say, the leader only reaching 10 out of 22 even with 2nd votes. My intention was to just call it rather than delaying, but we could talk about it. For this time though, I'm calling it:

Shaquille O'Neal is Inducted at spot #8.

Nomination Vote

Kobe - 10 (trelos, ltj, rk, trex, ceoofk, ShaqA, f4p, AEnigma, speel, Ohayo)
Dirk - 2 (OaD, Dr P,)
Mikan - 4 (beast, eminence, cupcake, Moonbeam)
Oscar - 2 (Doc, Narigo)
Robinson - 1 (iggy)

Kobe Bryant is added to our Nominee list.

Close vote, exciting!

I definitely don't think we have to have a majority vote no matter what. Everyone has a chance to vote strategically with their alternate. As long as someone has the lead after counting out second place votes between the top 2 candidates, we should just move on. Now if we were tied after second place votes, then it would make sense to do a run-off with the people who didn't choose either choice.

[Moonbeam]

Immediate Hall of Fame RealGM post; you should make it into its own thread.

(And I appreciate having more material to use for my eventual Bob Lanier nomination. )

Are the excised rankings just players who played too few minutes to come across as notable? Might be appropriate to set a certain minimum games played threshold; I notice top rookies seem to do inordinately well.

Thanks! I might make a separate post at some stage once I have more data to share.

As far as the players listed, I've only included the ones who won awards within this timeframe just for simplicity (and character limits). If I made the data more public, I'd include everyone.

These figures will place an extraordinary amount of weight on players who miss some seasons within the window, so not only rookies, but players whose careers end as well.

Question for clarification - are you running each 5 year regression individually, or running a bigger one across the 80s and applying it to 5 year windows? (So are the box-score variable estimates the same in each one, or changing).

Yes, I'm running each 5-year regression individually, so the box score variable estimates are changing with each window. This is why it wouldn't be appropriate to compare the value of the Ridge/Lasso/ENet scores across windows alone --- each time, the amount of shrinkage that is performed is chosen by cross validation of that data set to produce the least error. In other words, a Ridge score of 1.5 in one 5-year window isn't necessarily comparable to a Ridge score of 1.5 in a different 5-year window.

I've been thinking about how to best present the results, and percentiles may be the best way to do it for that reason.

Very cool stuff! Do you have a Github? Would be very interested to know how you are regressing box-score stats on WOWY. Is it season total box-score or per game (average) stats, similar to basketball-reference BPM calculation? What is the WOWY dataset that is being used?

Also, #107 Michael Jordan: 1.929 for 1987-1991 Regressed WOWY, what do you make of this?

I should get around to getting a Github. I'm using R to perform the regression using a simple linear model for the raw scores, and then using the glmnet package with cross validation for the others. If there's interest, I'll write something up in RMarkdown and share the code I've written for this.

As for Jordan's score, I'd wager that the other estimates (Ridge, Lasso, and ENet) are better equipped to filter the noisiness of the data and would generally be more accurate. But beyond this, I'd say that WOWY data is at best a rough approximation of +/- data. It's as good as we've got prior to 1997, so it's something, but I'd caution against making as strong of conclusions based on this sort of data as you might for the same approach applied to +/- data. I'm going to have a look at comparing RAPM percentiles vs the RWOWY-Ridge estimates over comparable 5-year windows just to get a sense of how much they line up, but I haven't done that quite yet.

I've been playing around with box score data to run regressions on WOWY. Traditional WOWY statistics ignore what other players may be missing, so running a regression should offer an improvement. I've been looking at 5-year windows and have applied some shrinkage techniques like Ridge (as used in RAPM), Adaptive lasso and Adaptive Elastic net. I've been thinking about this with the view of voting for Magic here and thought I'd share the data I'm looking at before writing up my voting post. For simplicity now, I'll call this metric "RWOWY" for regressed WOWY, and then "RWOWY-Ridge", "RWOWY-Lasso", and "RWOWY-ENet" for the ridge, adaptive lasso, and adaptive elastic net variants.

These values make use of all games (regular season and playoffs) and all players who averaged at least 18 MPG for the season. I'll print out the results for 5-year spans over Magic's career for players who won at least one award (All-Star, All-NBA, All-Defense) during that span.

Immediate Hall of Fame RealGM post; you should make it into its own thread.

(And I appreciate having more material to use for my eventual Bob Lanier nomination. )

Are the excised rankings just players who played too few minutes to come across as notable? Might be appropriate to set a certain minimum games played threshold; I notice top rookies seem to do inordinately well.

Agree, I think a separate thread where people can discuss salient players in the data would be great.

I'll look to make a separate thread at some point. At this stage, I have box score data from 1977-2022 (my code spat out an error for 2023 so I'll have to fix it). I'm now getting some pre-merger box scores as well. Do you (or anyone else) think there'd be interest in this before all of the data would be ready?

[LukaTheGOAT]

Spoiler:

2015 finals


2016 finals


2018 WCF


2019 2nd round Rockets


2022 Western playoffs


okay, but these really aren't that great of numbers. every single one is scoring in the mid-20's with a dip to the low 20's in the 2016 finals, with nice but hardly amazing 4-6 rebounds and assists. if this was the 1998 finals and the game was being played at a pace of 76, these would be really good, but much less so in the modern era. and the TS% for almost all of these is in the mid-to-high 50's.

you seem to want to say that the TS% still looks good compared to a normal player (certainly not in some of the ones i quoted), but part of the steph curry story is that he is a TS% god and he has amazing off-ball gravity. when we compare him to other players, we're working from that baseline. same as we assume hakeem is a great defender and iso scorer or lebron is a great scorer and passer. if one of curry's main things goes away, and not only falls off from "singular greatness" to elite, but starts falling into the territory of barely above average, then we're taking away one of the big things that makes steph so great.

just saying "he has impact beyond the box score" isn't a get out of jail free card. for one thing, these other players at this level also have impact beyond the box score. magic and bird and garnett (and guys voted in before like hakeem and duncan) all have things that don't show up. and another, his impact beyond the box score isn't infinite, able to swallow up any decreases in his play. or at the very least, if you're telling me that curry with "meh" box score stats compared to other all-time greats is still just as good as them, then what is he like when he's actually shooting well and putting up big numbers? is he the best ever on the days he's putting up 33/8/7 like against OKC? people are already nominating him as high as 8th, presumably with a longevity penalty, so let's bump that up to 5th or 6th with more longevity. and that's with subpar box scores. when he plays well does that bump him up to the best ever? because that's essentially what's being implied if he's ranked this high with those box score numbers. it seems much more likely that when his box numbers take a big hit, we shouldn't just assume that his non-box impact somehow increases to offset it and that, much more likely, he just isn't as good in those situations.

and you say if this is his worst, then he must be great. but it also seems to be about his best. it's not like i've cherry-picked random series from steph's career. these are the most important ones. from his peak. i didn't just go find some 1st round series against an 8th seed where he was bored and didn't put up big numbers. i didn't include series from 2013 or 2014. i didn't include the 2018 finals, which were weirdly inefficient, because it was a cakewalk and i doubt he felt threatened enough to do any more than he did. i picked most of his most important series from his peak years. and none of them are standout. 23/5/4 on 54 TS% in the year 2019 is just bad. chris paul almost had the same game score and he looked so washed that that offseason they had to add 4 picks just to trade him for russell westbrook. 26/6/5 on 60% TS% in the west playoffs in 2022 is nothing to write home about. steph put up 30/6/5 on 59 TS% this year in the playoffs, and his PER was only 20.4! WS48 of 0.131. not even remotely impressive compared to some of the guys we're talking about. that's how much the stats have changed and why a nice 25/5/5 line isn't what it used to be, especially for someone whose case is built almost entirely on offense.



i didn't ignore either. without these, steph's playoff career would arguably be almost entirely devoid of big moments in high leverage situations. as it is, the last 3 games against OKC and game 4 against boston lifted him up a decent number of spots in my rankings, as i think it showed separation from some of the guys i didn't see do that. i wouldn't have him in my top 15 without those moments, much less arguing maybe 12th or 13th in this project.





in order to have much hope? if 2 plays go differently at the end of game 7, his team wins the title. that's a lot of hope. playing well twice and almost winning a title is incredibly fortunate. plenty of guys not only have to play great to get a win, but even have some great performances thrown away on losses. not just 4 good games that are wins and 3 bad games that are losses.

Curry's relative true shooting is still elite even if not singularly outlier.

During his 3-year peak from 15-17, he has a rTS% of 8.6% on Adjusted 28.9 pts per 75.

If you want to look at his 5-year Peak from 15-19, he is at rTS% of 7.3% on Adjusted 28 pts per 75.

That's elite.

but that's why i'm bringing up his biggest series. and i never said he wasn't efficient in 2017, just that it was easy. he can put up a 72 TS% against the 2017 spurs after kawhi is injured or smack the crap out of 2019 portland, but those aren't really where his team's title chances were up for grabs. these other series were.

and i would be more sympathetic if he was just facing defensive juggernaut's in these series, but he wasn't. the 2022 celtics were the first defense he ever faced below -4.2 rDRtg, and the -4.2 team was all the way back in 2013 against the spurs. so none during his peak (not counting the kawhi-less spurs for obvious reasons). the 2016 cavs have a bad defensive point guard and an exploitable slow-footed big in kevin love. the 2018 rockets had a very good defense but it was still just ranked #6. the 2019 rockets weren't quite as good. the 2022 western opponents weren't amazing i don't believe, with denver being pretty bad. 2015 cavs were really good at defense, but that seems to be the only one in the series i mentioned. if he was going up against 2004 pistons/2008 celtics level teams and struggling then ok, that's understandable. happens to the best of 'em. but he was seemingly given prime opportunities to dominate and time and again ends up in the mid-to-high 50's on TS%.


also, just so i know, is rTS% in the playoffs (i assume your numbers were for the playoffs) based on comparing to the league's TS% in the regular season or the TS% in the playoffs? and does it account for things like being on one of the best defensive teams that plays 4 rounds and hold down everyone else's TS%?

The relative TS% takes the player's TS% against the team minis average TS% allowed for the opponent. So it does help to account for the quality of the opponent defense.

Steph had a rTS% of 4.5% in the 15 Finals.

Had a rTS% of 8.8% in the 16 WCF.

Had a rTS% of 5% in the 16 Finals.

Had a rTS% of 3.3% in the 18 WCF Finals.

Had a rTS% of 5.5% in the 19 Finals.

His efficiency goes down, but I think you see that even with the best stars when their team is challenged more.

[AEnigma]

I'll look to make a separate thread at some point. At this stage, I have box score data from 1977-2022 (my code spat out an error for 2023 so I'll have to fix it). I'm now getting some pre-merger box scores as well. Do you (or anyone else) think there'd be interest in this before all of the data would be ready?

I think there would be strong interest because people are naturally impatient, but for the purposes of this project you have largely covered the most immediately valuable chunk (1977-93 is where we have the least easily accessible data, so therefore these regressions add have the most insight to offer).

Other personal projects such as this have been released on rolling bases, but I personally have no issue waiting for it to be “complete”. Prevents rampant speculation “oh I am sure so and so just needs these years to kick in”. However, I will preemptively note that in recognition of that 1977-93 timeframe, people may be interested in how a player like Julius Erving seems to fare within the next couple of weeks.

[LukaTheGOAT]

Read on Twitter

Anybody know what his updated CORP numbers look like if he has posted them? It seems with the addition of 2019 and 2022 that Curry might be in this top 16.

He did the new rankings in a podcast format. Scroll to the very end

https://thinkingbasketball.net/2017/12/11/the-backpicks-goat-the-40-best-careers-in-nba-history/

[WestGOAT]

I should get around to getting a Github. I'm using R to perform the regression using a simple linear model for the raw scores, and then using the glmnet package with cross validation for the others. If there's interest, I'll write something up in RMarkdown and share the code I've written for this.

I'll look to make a separate thread at some point. At this stage, I have box score data from 1977-2022 (my code spat out an error for 2023 so I'll have to fix it). I'm now getting some pre-merger box scores as well. Do you (or anyone else) think there'd be interest in this before all of the data would be ready?

I would definitely be interested in reading up on the code (ChatGPT is a godsent for this ), as well, as the data there is so far!

[homecourtloss]

I've been playing around with box score data to run regressions on WOWY. Traditional WOWY statistics ignore what other players may be missing, so running a regression should offer an improvement. I've been looking at 5-year windows and have applied some shrinkage techniques like Ridge (as used in RAPM), Adaptive lasso and Adaptive Elastic net. I've been thinking about this with the view of voting for Magic here and thought I'd share the data I'm looking at before writing up my voting post. For simplicity now, I'll call this metric "RWOWY" for regressed WOWY, and then "RWOWY-Ridge", "RWOWY-Lasso", and "RWOWY-ENet" for the ridge, adaptive lasso, and adaptive elastic net variants.

These values make use of all games (regular season and playoffs) and all players who averaged at least 18 MPG for the season. I'll print out the results for 5-year spans over Magic's career for players who won at least one award (All-Star, All-NBA, All-Defense) during that span.

Immediate Hall of Fame RealGM post; you should make it into its own thread.

(And I appreciate having more material to use for my eventual Bob Lanier nomination. )

Are the excised rankings just players who played too few minutes to come across as notable? Might be appropriate to set a certain minimum games played threshold; I notice top rookies seem to do inordinately well.

Agree, I think a separate thread where people can discuss salient players in the data would be great.

The Top 100 project keeps on giving with posters such as @Moonbeam and others giving us new pieces of data. It’s interesting seeing the evolution of the project and the threshold of what is considered viable “proof” for a claim and/or position today vs. in early 2010s or before.