Alright, going to start with Steph. For those who are interested, Hakeem stuff can be found at the bottom.
lessthanjake wrote:OhayoKD wrote:Yeah, so I think you're interpreting some of this data very weirdly. Let's start with RAPTOR. A statistic that is less predictive than LEBRON or RPM because it does not input RAPM directly and it's creator was working on the assumption that perimeter players were overrated and two-way bigs were underrated.
You can come up with reasons to discount specific measures if you want. I have provided the results from an enormous amount of measures
Indeed, examining the methodology, blindspots, uses and effectiveness of metrics is something we can do and probably --should-- do. RAPTOR is demonstrably worse at predicting results and predicting results
as rosters change because it does not input "winning" as directly as metrics with similar tech. It also is set with priors that skew towards pg's and two-way wings relative to what was available(that was literally a reason for constructing the metric according to nate silver). All that should play a factor in what you extrapolate from the results. As it would would with any other sort of metric. I also feel I have been open about uncertainty, but when you are making decisions that is something you have to deal with. There is a difference between acknowledging uncertainty and trying to stack the most metrics because they're all not perfect.
Of course, this is all somewhat academic because only one of these stats potray Steph as the impact king of the last 10-years(it's not RAPTOR).
To begin with, we do actually have metrics from Steph’s prime that are not rate stats and do still have Steph easily ahead. For instance, RAPTOR is a rate stat, but they also convert that into a non-rate stat with their WAR measure. And by that measure, Steph is ahead of LeBron every year in the last decade except 2017-2018 and of course 2019-2020.
Very cool. Here's the thing.
Steph does not do the best in RAPTOR WAR. Credit for beating Lebron in a set of data that excludes all his MVP years(we've went from "half" a prime to "barely any of it"), but both James Harden and Nikola Jokic come out at 1st just as many times(
3), and Jokic sees a bigger gap than any of Steph's
twice. Steph only gains significant seperation in
2016 with
Draymond finishing second.
LeBron was barely in his prime after 2012-2013? I’m confused. Because that’s definitely not consistent with arguments used in LeBron’s favor (and I think correctly used in his favor). It seems a bit selective to now narrow down LeBron’s prime to be only a few years, and that’s not something I think the vast majority of people would agree with.
The arguments in his favor largely center on 2015-2017, and 2020 with the other years specifically getting a playoff focus. From an impact perspective 2014, 2019 and 2021-2023 are all nadirs with 2018 specifically included in prime largely on the basis of the playoffs. From these type of metrics even 2005 is eligible for league-leading. This is also partially box-informed. Even the prime-years do not look like "prime" years from box. These sort of metrics are really about
replication and frankly you are probably fixating too much on "gaps" as this stat was designed with cross-period comps in mind. These sorts of metrics are good for establishing baselines: if a player is posting scores at the very top of the data-set(not just one year) over and over significantly more than anyone else...then they are most likely the most valuable. These stats do not really track who is fluctuating up(peaking) highest. That is why you need to also be doing real-world analysis. The scores on these scales
are not a substitute for tracking what actually happens. But we can get to that later, because, regardless of approach...
And yeah, Steph is behind Jokic the last few years in most of these measures (as is everyone else). But Jokic is playing at an unbelievable level the last few years, and Jokic wasn’t finishing ahead of Steph earlier in the past decade. I’m not asserting that Steph has been at the top of these metrics every year for a decade. No one has.
You specifically argued Steph has been "the impact king". That doesn't work how you've been interpreting data if he is not winning against a contemporary
both in frequency,
and in terms of
gaps. Jokic, not Steph is the "impact king" of this 10-year period going by box-informed WAR using
your approach. Jokic was drafted in 2014.
And of course, this isn't really just impact. Much like how you decided to look at what MJ's 90's scores looked like without JE's prior(is he top 5?), there's merit to looking what the non-box(iow, not skewing towards an
assumption of what impact looks like) and if we were to do that...
Things go from bad to worse if we switch to the RAPTOR on/off, because now, somehow, "prime" Lebron
looks better, coming ahead several times including
2016 and 2017 and finishing first just as often(2). Steph is only spared the embarrassment of Lebron coming out at first a 3rd-time thanks to 2016-leader
Draymond Green. He could not carry Steph past Lebron in the finals, but he got it done on the illustrious stage of 538:
RAPTOR on/off is not an independent metric. It is an input into the overall metric. And you can see why it’s not the overall metric because it is clearly super super prone to randomness (for instance, Nic Claxton ranked #4 in the NBA this past season). But yes, if you want to say that LeBron outdid Steph in one particular component that goes into an overall metric that Steph does better in, then I guess you can do that.
It is the impact component of RAPTOR and when it repeats what most impact analysis does, I'm not sure you should be using non-impact stuff for a case that Steph was the "impact king". Regardless of approach though, Jokic takes that mantle from steph pretty clearly. And if we focus on impact, Lebron is still beating him right through what would be his wizard years.
The most predictive/stable metrics like say LEBRON seem to hate Steph more:
You’ve elsewhere listed EPM as one of the metrics you think is most predictive/stable. Why is that dropped here? How does Steph fare compared to LeBron in that measure? They even have a non-rate-stats version of it. Who has done better in it more often?
It's dropped because we only have not-behind-a-paywall access to 2023 data where Steph edges ahead of Lebron but is well off #1
2023 stuff favors Lebron otherwise(LEBRON, RPM, raw-stuff, wowy, lineup-ratings ect) with the exception of RAPTOR(on/off and box favor steph). Neither are #1(or particularly close) though.
And even with the metric you chose here (i.e. LEBRON), which one of them has been above the other more in the 9 years of Steph’s prime? Is it not the case that Steph has been above LeBron 6 out of 9 years in the default version of the stat (i.e. the rate-stat version) and 5 out of 9 years even in the non-rate version?
Yes though I do not think felling 19, 21 and 22 lebron makes you an impact king and obviously this swings back to Lebron depending if we define steph's "prime" at an earlier point.
There is no point doing a non-box filter here because even with box Steph does not come close to an "impact king" leading the league only twice by less than a win in 2015 and 2017. He falls behind Lebron in 16 and 18 and is just a win ahead in 2017. For the four-year stretch of 14-18(taking Lebron to the spot where most top-10 candidates are at or near retirement), Lebron generates more wins.
Again, we can’t compare year to year stats like that.
Um, we can't? These types of metrics have multi-year filters iirc, and the creators do not seem to have any issue with comparing different years to each other which may suggest they were actually designed for that to be viable. As is, Lebron has 4 scores higher than any of Steph's and 3 of the top 4. And while this probably changes if there was a way to isolate out the box-stuff, even KD looks better with 3 higher scores. Interestingly enough box-LEBRON actually likes steph better(his years move up) than actual LEBRON does though Lebron, Giannis, and Jokic still top him.
Regardless, using
your approach Jokic is the actual impact king of the last 10 years with 3 leads and 3 bigger gaps. James Harden also looks better than Steph with 3 leads and 2 bigger gaps. Steph is not an impact king here and steph going up with a box-only filter would suggest the "impact" bit of this would rate him even lower.
Finally there's RPM....
lessthanjake wrote:
Okay, but the size of the gaps depends on the metric we’re looking at.
For instance, if we look at RPM, Steph had two out of the three largest gaps to 2nd place that Steph and LeBron have had (including having the year with the largest gap)
Yeah, I don't think that is correct:
Steph has the 2nd and 3rd biggest gap, but Lebron has the 4th, 5th, and 6th biggest gap(2013, 2009, 2007), and led the league
9-times to Steph's
4. Lebron seems rather clearly advantaged to me there and things probably get lopsided if we look at larger stretches. Not to mention the scores here being artificially suppressed...
So, when you went to get those screenshots, did you notice how when you went to the webpage for RPM, it defaulted to ranking by “RPM” rather than ranking by the “WINS” category, and that to get your screenshots you had to specifically change it? Did you check whether what I said about RPM was accurate using the actual default version of the stat that is actually called RPM? It was.
And, if you insist on using the “WINS” category instead, please tell me who finishes ahead of the other more often in the last decade.
Noted.
For the last decade? Steph. And here(and only here) Steph has a claim as an "impact king" over this already small time-frame as Jokic does not do so well. It's not the 5-year-run(6-out-of-8) we see from Lebron, but if you want to say he ran RPM for a decade, go for it.
Here's the thing. The higher APM scores are artificially suppressed and at a certain value treshold(like say, the biggest raw delta in with/without and before/after in nba history) a metric like RAPM tends to misallocate value to role-players(especially over 1-year increments). In 2010 Lebron scores first with the second highest score in the database(Dirk's 2011 is #1). #2 is Anderson Varejoa, a player who with 28.5 minutes per game, notches one of the highest scores in the set, to join the pantheon of one-off adjusted-plus-minus toppers alongside 2009 Lamar Odom.
Did you ever wonder whether maybe other players on Steph’s team have the same effect in some metrics? Draymond, perhaps? Or perhaps others? You’re not talking about a LeBron-specific phenomenon.
I do. But then I remember Draymond
-> has repeatedly replicated top-tier impact signals
-> played more minuites than him at his peak(per-game too in 2016)
-> has played more playoff minutes(even on a per-game basis) several times
-> has won a playoff series without him or durant(may have won another)
-> and literally outimpacts steph over extended stretches in the playoffs
-> has outplayed steph in the finals
-> has outplayed Steph over a playoff-run
But you know what? Why don't we check. Just in case...
Yeah shockingly, not at all the same. Warriors are as good in dray no steph minutes as they are in steph no dray minutes and notably,
they've played the same amount of minutes without the other. As a kindness I'll leave out the playoff minutes but as one might expect, Draymond and Lebron look
better. Steph and Andy look
worseI’ve provided a bunch of playoff and regular season + playoff metrics in my posts. Steph looks good throughout. Take a look at the “spoiler” section on impact metrics in my prior post, as well as that post’s listing of where Steph ranked in the league in PS AuPM/g each year. It looks very good for Steph.
Steph falls behind both Lebron and Draymond. He is not even the undisputed impact king of his own team. I also do not know that a period with prime cp3(an all-time impact darling), kobe, dirk, nash(impact darling), and wade could properly be described as a lull. Depending on what you use, Lebron also takes the lead facing late-prime Duncan and Garnett.
I’d direct you to my earlier post, which has a “spoiler” section that addresses this completely. The upshot is that to the extent Draymond has an edge in low-sample-size playoff data, it’s a result of early-round playoff series’s against weak opponents that Golden State easily beat. Once you look beyond those, the Warriors do better in the playoffs with Steph on the floor than with Draymond on the floor. I’d hardly put much credence in a fact that is based on a low sample size of data that is reversed once you eliminate the data points that are clearly the least important.
Fair I suppose. Granted Draymond did not have the benefit of playing with Steph while Steph benefitted from playing with Dray, but we can put it down to competition if you want.
That said, the standard here is "impact king", not "looks good". Steph doesn't rank 1st once and is also offering less value than his teammates in 2016 and 2018 because of missed time. Moreover, AUPM is a combination of on/off and a box-aggregate, the former being something Steph is likely getting inflated by way of colinearity...
https://forums.realgm.com/boards/viewtopic.php?p=106319069#p106319069...and the latter being something we have now have repeatedly seen(even in this post) typically elevates Steph relative to impact-only data.
Still a strong case as the 2nd-best player over a 5-year stretch or so, but an impact king that does not make. Nor does it really give him a notable advantage over the field.
As for Hakeem, the potential victor of round 6(I believe this would be the highest he's ever gone?), I'll make a final foray into the plus-minus battleground.
First a bit of house-cleaning:
Doctor MJ wrote:lessthanjake wrote:I feel like some people here must already be aware of this, so it’s probably redundant information for some, but I did just stumble across RAPM data for 1993-1994, 1994-1995, and 1995-1996, using Pollack’s data. See the links in the below:
https://www.apbr.org/metrics/viewtopic.php?t=86711993-1994 RAPM:
https://docs.google.com/spreadsheets/d/1VSPxw_RVZ-WBOHM5J434efmLjMYGVYi8CrA59WS7GuA/edit#gid=20183146841994-1995 RAPM:
https://docs.google.com/spreadsheets/d/1hnhhuQhlY-hp0Qt6xT-cUe2tqBF3TxPjYW8g0mblnmQ/edit?pli=1#gid=19325033021995-1996 RAPM:
https://docs.google.com/spreadsheets/d/1nqWP4Lu7lBjAydga0vrN1Gv4L686iA1AktvFj4vgKSo/edit?pli=1#gid=1715147358What I find particularly notable for these purposes is how well Shaq does. Shaq is ranked 8th in 1993-1994, 2nd in 1994-1995, and 7th in 1995-1996. This actually compares pretty favorably with Hakeem, who is 4th in 1993-1994, 6th in 1994-1995, and 11th in 1995-1996. And this is Hakeem’s peak, while it is just the very early years of Shaq’s prime. Hakeem is also behind David Robinson and Karl Malone each of these years (and of course MJ in 1995-1996). Pippen was above Hakeem in two out of the three years as well, though Hakeem was ahead of Pippen by a lot in the other year.
Of course, Hakeem’s playoff resume in these years was great, and I don’t think this data includes playoffs, but it still does seem worth noting that peak Hakeem was still being outdone in RAPM by a few players, including one current nominee who wasn’t even in his peak years at the time.
So, unless I'm mistaken that's not actually RAPM, because to do RAPM you need access to lineup data from minute to minute.
Rather, what's being done is a regression based on the raw +/- totals for the entire season. It's worth doing, but it's not what "APM" means in basketball statistics.
And yeah, the thing with Hakeem is the playoffs. If there were no playoffs, Malone & Robinson would be ranked higher than him.
It also excludes 2 Hakeem years that would potentially score higher than any of 94-96 from an "impact" perspective in 93 and 92 where the Rockets were
-11.2 without and
-0.2 with over a decent sample(12 games per szn) one year and then jumped to
+4.5 with Hakeem playing every game the next. We are comparing "early-prime shaq" to what should be "late-prime" Hakeem and "early-prime Shaq" posts some of his best signals. IIRC, per Ben Shaq was, on average, worth something like 18 wins in Orlando. Backpicks is down and i do not recall the exact numbers or time frame, but from memory it's on par with any non-magic rs stretch through the 80's.
1994 in particular looks like a top 60 signal from the last 30 years. Considering that the 92 Rockets were outscored by 10 points in games without Hakeem, it's not hard to see inclusion of 92/93 giving Hakeem a top top 3-year signal.
All considered, I don't really think this looks bad for Hakeem, and again, there is plenty placing him as a top-10 worthy
regular-season candidate:
https://forums.realgm.com/boards/viewtopic.php?p=107696943#p107696943(first spoiler)
The
largest sample of "off" for Hakeem comes in 84 and there...
...yeah Hakeem looks pretty good.
Olajuwon, considered a
rawer prospect than his draft-mates who would need longer to flourish,
immediately sees an mvp-ish worthy delta with the defense spiking by 2-points. Hakeem's teammate finishes ahead of Olajuwon in MVP voting and yet the Rockets are unaffected when that teammate misses half the season in 87. All three of those years Hakeem's team, similarly bad before his arrival, wins more than his far more lauded draft-mate(finishes 2nd in mvp voting despite less wins than olajuwon managed with a coked-out roster), and he hits another level in b2b2b playoffs as the Rockets dramatically overperform.
If we look to other large samples:
mpact was muted early on because of this. There's only so much re-examining one can do here without firing up every Rockets game from the 80s and hand-tracking on/off and everything else.
Then again, some of the other evidence suggests Hakeem wasn't as valuable as the eye test would indicate. That Houston wasn't as lost without him and that distributing what he did across the team wasn't impossible. (Same can be said of Duncan.) We have such a tendency to de-emphasize teammates and coaching as years go on and only remember co-stars...of which, Hakeem basically had none.
I do know that in 1986, in his 2nd year, Olajuwon missed 14 games. Houston was -0.8 per contest in that time, which was
7.3 points per game worse than in the 68 games Hakeem played (+6.5 in those games), improving both on offense and defense.
I know that in 1991 -- a notable down year for Dream -- in the 26 games he missed Houston
outscored it's opponents by 2.4 ppg (16-10 record). They were only slightly better with him
(+4.0 overall) but while the defense improved about 3 points the offense regressed slightly.
I know that in 1992 -- another "bumpy" season -- in the 12 games he missed Houston was outscored by 10.8 points per. The DRtg was a dubious 117.2. With Dream, they were -0.2
(+10.6 difference).
I know that in 1995 when Hakeem missed 10 games,
Houston was -4.3 without Olajuwon and +3.0 with him. Again, the offense was better, but the defense fell apart in his absence. (The DRtg was 116.8 without Hakeem.)[/quote][/spoiler]
Again, pretty good. So good in fact, the blurb ends with
Personally, I settle in the middle on Olajuwon, with his career ups and downs. But I do question if that's wrong. My instincts tell me if he had a better situation (eg San Antonio, 1998!!!) that we might regard Dream as a serious, serious GOAT candidate. Yes, even challenging Jordan.
To his credit Elgee
acknowledges uncertainty. Depending on how you interpret the data you could also get Hakeem pretty low, but step one is acknowledging that there is different data that points to different things. Once we get there, stop two is applying what you know, both about the data(sample-size, methodology, which years might look better or worse, ect), and about the player(skill-set, historic trends, year-to-year context, "production", film, ect).
Since the larger samples are more favorable to Hakeem, Hakeem is very similar to players who look exceptionally valuable across a variety of emperical approaches(Duncan in general, kg, ect), and Hakeem's teammates do not seem to have a big effect over sizable stretches(thorpe, Sampson), I think taking a higher interpretation of what is available is reasonable. And then if you take such an interpretation, that leaves us with unrivalled playoff-elevation and unusual longevity while playing in a for from optimal context. Put it all together I think you have a solid case for top 6 and maybe even higher(I had him 5).
Regardless, confidently asserting there is "no impact case" seems to require a very selective lens of what qualifies. One that I don't think holds up so well when standards are applied consistently.
