What impact metrics show MJ as a GOAT candidate?

[MyUniBroDavis]

Wait I don’t understand so wouldn’t all pre 97 measurements be estimates of impact measurement using only box score components to try to account for the missing impact component?

They are trying to pull a fast one on us. They cite PIPM for LeBron which is a Box Score and Plus/Minus blend then cite PIPM for MJ acting like it's the same stat. When in reality, PIPM for MJ is all box score. You might as well cite BPM because it's the same exact stat as PIPM for pre-1997 players.

What u mean us I was just asking a question lol

My issue with a lot of plus minus estimates using pure box score data is it’s based on an assumption that the same play style is effective pre 97 to post 97, but the defensive environment changed a good amount and then the offensive style did too so idk if that’s a good assumption. Beyond that there’s gonna be a lot of error at an individual level

[colts18]

Wait I don’t understand so wouldn’t all pre 97 measurements be estimates of impact measurement using only box score components to try to account for the missing impact component?

They are trying to pull a fast one on us. They cite PIPM for LeBron which is a Box Score and Plus/Minus blend then cite PIPM for MJ acting like it's the same stat. When in reality, PIPM for MJ is all box score. You might as well cite BPM because it's the same exact stat as PIPM for pre-1997 players.

What u mean us I was just asking a question lol

My issue with a lot of plus minus estimates using pure box score data is it’s based on an assumption that the same play style is effective pre 97 to post 97, but the defensive environment changed a good amount and then the offensive style did too so idk if that’s a good assumption. Beyond that there’s gonna be a lot of error at an individual level

I'm not referring to you. I'm talking about the poster here citing PIPM for LeBron and MJ without realizing that it's not the same stat.

[DraymondGold]

They are trying to pull a fast one on us. They cite PIPM for LeBron which is a Box Score and Plus/Minus blend then cite PIPM for MJ acting like it's the same stat. When in reality, PIPM for MJ is all box score. You might as well cite BPM because it's the same exact stat as PIPM for pre-1997 players.

What u mean us I was just asking a question lol

My issue with a lot of plus minus estimates using pure box score data is it’s based on an assumption that the same play style is effective pre 97 to post 97, but the defensive environment changed a good amount and then the offensive style did too so idk if that’s a good assumption. Beyond that there’s gonna be a lot of error at an individual level

I'm not referring to you. I'm talking about the poster here citing PIPM for LeBron and MJ without realizing that it's not the same stat.

Hi colts18! If you're referring to me, I already stated the explicit differences between the two PIPM versions in my first post in this thread (https://forums.realgm.com/boards/viewtopic.php?p=103518100#p103518100), and have already further addressed the concerns you mention in my later post (https://forums.realgm.com/boards/viewtopic.php?p=103530610#p103530610)

So if you are referring to me, I'd suggest replying to one of those, rather than ignoring the points I've already raised. If you're not referring to me, my sincere apologies! Keep calm and carry on

[DraymondGold]

A. Intro and ‘What is GOAT tier’?

Seems people here have been able to stay mostly respectful, so I'll take that sign of progress.

I appreciate this OhayoKD

Before going ahead, let me just start out by saying I'm hoping we can avoid turning this into another Lebron vs Jordan thread. With that in mind, my personal focus here is not whether these metrics portray Jordan as the clear-cut best GOAT player, but whether they portray him as a GOAT candidate. Do they merely portray him in the Top Tier of players for peak/prime.

I appreciate this sentiment but if this was your objective, I'm surprised then that you almost exclusively focused on the Lebron stuff. I think this part in particular is relevant when we are talking about of all time tiers:

Agreed, but it's a bit of a bind -- I can ignore all your (and others') comments about LeBron to push the discussion that way, but that would also be poor discussing. The other issue is LeBron's the only standard Mount Rushmore candidate (Russell, Kareem, MJ, LeBron) who's fully in the impact stat era, so if I'm trying to compare Jordan to to other Mount Rushmore candidates to show he's in a similar tier using statistics, I'm pushed towards using LeBron for comparison.

I guess that's all to say, I'm trying not to focus too much on whether LeBron is actually better for prime/peak, just whether Jordan is close enough in prime/peak to be in the same tier (since LeBron is presumably in Tier 1 all time). And if there's other places to shift the discussion towards Russell or Kareem or whoever, I'm open to go there instead.

Keep in mind we don't have the data for players like Kareem, Hakeem, Bill, Magic, Bird, Walton, Wilt, or Russell. Jordan is competing in a very, very narrow field here and still doesn't look the best.

Saying Jordan ranks 3rd or 2nd behind other top 10 candidates in metrics that exclude most of Jordan's contemporary and historical competition doesn't really work as proof he's "GOAT tier" unless we are using Colt's much, much broader standard. One you don't seem to follow considering that you have advocated for the exclusion of players for Duncan and Hakeem. This applies, I think to your assertion of AUPM as definitive proof(more on that later).

You’ve suggested Top 3 in a post-1997 stat isn’t high enough to be considered GOAT-level.

Here’s the problem: there is no absolute-consensus #1 player across all impact metrics, not since 2014, not since 1997, not since 1977, not since 1955, not since 1946, not in peak, prime, or career. There just isn’t.

So if being Top 3 in a post-1997 stat is a disqualifier from being GOAT level in impact statistics, that leaves us with two options:
1. Nobody is GOAT tier in impact stats (since nobody, not one, is Top 2 or Top 1 across *every single* impact metric)
2. We need to lower the threshold for what’s considered GOAT level in impact metrics below Top 2 or Top 1 only

If you favor #1, that’s perfectly fine! The boundaries between what we consider “all time” and what we consider “GOAT tier” can be left up to personal preference. And if you prefer to set it higher than any player has gotten so far, that can be perfectly consistent with your criteria! May Victor Wembanyama finally be the one

But me personally, I prefer to have a handful of players in my GOAT tier. So I prefer #2. To me, it’s okay to be “only” Top 3 all time in a stat we have in the time ranges those stats have and still be GOAT level, so long as you’re consistently near the top (hopefully with lots of Top 5s, Top 3s, and Top 1s) across the array of metrics we have. MJ fits this description, so I consider him GOAT tier statistically.

B. Do box-inputs to plus-minus stats and Box-based stats overrate Jordan?

Additionally, when you are dismissing things that are directly drawn from winning like on/off(that directness is very much the point of impact analysis) as useless while championing crude approximations because they make "corrections", it's probably relevant to consider when these corrections are actually making the data more inaccurate:

In short sample sizes, the box-input corrections to actual on-off-based stats are measurably making things more accurate. The idea that raw plus-minus incredibly noisy is complete consensus throughout the community. Some stats aren't even fully stable in full-regular-season sample sizes. So what do we do with this noise? Well, we could try to apply context (which we should anyway!), but the issue here is we don't know for sure whether the noise is pushing the player's value up or down, or how much the noise is changing the player's value. So these box corrections (like AuPM) can measurably make us more correct in small samples (that is to say, closer to the large-sample actual plus-minus-based value that corresponds to contributions to winning).

As Lebron and Jordan are virtually tied on the offensive portion of all these metrics, simply replacing the defensive component with actual impact data, knocks Jordan off his perch. And remember, this is not including Kareem [streamable][/streamable]whose defenses were 4 points better, or Russell who won the most, by a landslide, on the strength of his team's defense.

BBR BPM is on par with RAPTOR in terms of predictive accuracy IIRC(notably behind direct rapm extraps. like EPM when tech is equalized). As far as BPM is concerned, Jordan is a significantly better defender than Kareem. Simply hedging between defensive impact signals and defensive box-score data knocks knocks Jordan out of range.

That a metric makes adjustments of some kind does not make it inherently better, and proper analysis involves weighing the merits and cons of different metrics and then deciding what adjustments/caveats/context needs to be applied. There is a trade-off here.

You get less noise, but you also get inaccuracy that skews towards a certain archetype. And when the "corrected" data is consistently disagreeing with "real" data, then adjustments should be made. That is the value of "raw" impact. And any credible impact analysis will factor in those types of signals.

The on/off Ben calced is, to my knowledge, the only available sample of data which doesn't utilize an artificial scale and accounts for defensive impact. Don't you think the idea that we shouldn't even consider this while we use metrics that equate steals per game with defense a little silly?

Good points here!

Absolutely, various metrics can tend to favor certain archetypes. There's the famous issue of earlier Basketball-reference all-in-one stats (either PER, ws/48, Basketball Reference's BPM 1.0, I forget which) consistently overrating bigs with assists and guards with rebounds.

Here, I definitely agree that box-only-based stats tend (but not always) to overvalue perimeter defense, particularly steals. I definitely don't have peak defensive Jordan over peak defensive Kareem in absolute defensive value (yay, a non-LeBron goat candidate to talk about! ). A few qualifiers that may shrink the defensive gap between Jordan and Kareem:
(1) One could make some sort of team-building relative-to-position argument in favor of Jordan... i.e. that Jordan's a bigger defensive outlier among guards than Kareem is among bigs, and thus teams get a bigger boost choosing Jordan (because it's easier to replace big man defense). Not sure this is very convincing, but at a minimum I do enjoy these sorts of philosophical discussions about how much to give 'bonus points' for scarcity of value at a position.
(2) I'm also open philosophically to the idea that Jordan's all-time motor leads him to 'get the most' out of his defensive value more consistently as he gets older. Kareem and LeBron both have seasons when they get older where their defense slides behind what they're capable of, maybe to preserve energy for the offensive end, or to preserve energy in the regular season for the playoffs. For example, I'd absolutely take 96/97 Jordan defensively over 2018 LeBron defensively. Kareem also has a reputation for lowering his defensive effort as he got older, though I'm less familiar with how his effort waxed and waned in specific seasons.

To shift our the conversation back to statistics (this thread's focus), a few more qualifiers about why we shouldn't throw out out the box-based statistics:
(1) Even if steals are overrated in box-based defensive measurements (e.g. if the box stats miss the fact that the steals come expense of unnecessary gambling, as they may for Jordan), steals are still individually the most valuable defensive play someone can make. A guaranteed change of possession and high probability of starting a fast break (the best type of team offense) is super valuable.
(2) Some of the better box-stats I'm using have input weights that are calibrated overall, rather than calibrated for offense and defense separately. In other words, we calculate those stats as a whole, then split into defensive and offensive components after. This may effect the error: in this case, a player that’s overrated on defense may actually be underrated on offense (since the mistake is whether we attribute the value to offense or defense, not in the overall value). Not a guarantee, just a possibility. Like you say, it may also mean they’re overrated overall.
(3) The fact that stats overrate certain archetypes is why I tend to like looking at a variety of stats across the board (e.g. including whatever information we have from stats that are almost all value-based like RAPM/PIPM, hybrid stats like AuPM, WOWY-based stats, etc.). So if a player looks pretty good across the board, e.g. if he almost always places in the Top 5 ever, that's a pretty compelling signal to me.

C. WOWY vs WOWYR: Jordan, LeBron, Russell, Kareem

I also think there are some basic inaccuracies here that we should address, acknowledging I was under the incorrect impression pre-97 aupm also lacked plus-minus data.

Happy to talk more! I certainly could be incorrect in my understanding in other areas

That being said...

WOWYR is just like Adjusted Plus Minus (APM) to raw plus minus. It corrects for the context

WOWYR = Adjusted plus minus. There are problems we get with raw plus minus. We can get similar problems with raw WOWY.

...uh, no

I’ll see your “uh, no” and raise you one “uh, yes”!

WOWY is short for “with or without you”.
WOWYR is short for “With or without you, regressed”

The process that changes raw plus minus data to Adjusted Plus Minus (APM) is a statistical procedure called a regression.

You’ll note that regression (which again is the “Adjusted” in Adjusted Plus Minus) is right in the name of ‘WIth or Without You Regressed’. It’s right in the name!

So again: WOWY ~ Raw Plus Minus. WOWYR ~ Adjusted Plus Minus.

You can read more about how WOWYR is calculated here: https://thinkingbasketball.net/metrics/wowyr/. You’ll note this is the original article that published WOWYR, and it mentions the parallel of WOWYR and APM in the first paragraph.
You can read more about how APM is calculated here: https://www.nbastuffer.com/analytics101/adjusted-plus-minus/
[Note: Since this is a confusion I often make myself, a regressed/ “Adjusted” data is not the same as regularized from RAPM. The first just corrects for the context of other players, to isolate for the impact of individual players. Regularized is when we try to correct for outliers / noise to make the measurement more consistent. The first is a must-do if you want to be sure you’re looking at the impact of an individual player specifically. The second may be beneficial, but there are different regularizing schemes and some of the outliers they correct for may occasionally be actual signal, not noise]

Instead of using results from lineups within a game (play-by-play data) like traditional APM, game-level plus-minus uses final scores from game to game for the players from that game. This allows for a historical, apple-to-apples comparison of per-game impact from before play-by-play was available (1997).

As I covered before, the "correction" is marginal. But frankly WOWY and WOWYR doesn't really make a difference here, because as long as you are using large samples, even corrected impact still has Lebron consistently looking better:

Before Michael, the 1984 Bulls were a 27-win team (-4.7 SRS) with an average defense and a futile offense that finished 5 points worse than the league (rORtg). Jordan immediately provided the scoring punch that they needed and Chicago improved to just above average on offense in his rookie year, with an overall improvement of nearly 4 points per game. In his second season, he missed a significant chunk of time after breaking his foot, then logged fewer than 20 minutes in each of his first six games back. Excluding those sub-20 minute games, the Bulls played 15 contests with Jordan at a 40-win pace (-0.3 SRS) that year.

The ’06 Cavs were even more impressive, thanks to a breakout year from James. With Ilgauskus and Gooden now accompanied by Larry Hughes (a moderate creator and inefficient scorer), the offensively-challenged Snow and two shooters (Donyell Marshall and Damon Jones), Cleveland churned out a 5.1 SRS when healthy (56-win pace) with a +6.6 offensive efficiency in 30 healthy games. A similar rotation ticked along at a 51-win pace in ’07 (3.4 SRS) in a larger sample, but the offense regressed to near-average, meaning the ’06 result was likely an aberration. (LeBron’s offense regressed slightly in ’07 too, likely contributing to the backslide.) Still, the period demonstrated that pre-prime LeBron-ball could buoy offenses while stuffing the court with defenders and a few shooters.
(The cavs were a -9 srs team before drafting Lebron)

Jordan also lags behind Kareem in larger(>10 games per season) samples, and Russell. and Hakeem. The most relevant part of WOWY vs WOWYR here is the inclusion of 82 game stretches in impact analysis. And I think you and I can both agree that full season samples can be very, very useful, even if there's noise to account for. The only thing that has Jordan comparing favorably(to lebron, not everyone in history) is 10-year data, but again, let's consider the sample in question:

What about the 8 years Jordan made an ECF (won 6 titles)? He missed a total of 6 games in those years. How can you talk about a stat when there is literally no sample at all during Jordan's prime? We have just one season at all where Jordan missed significant games. His 2nd season in 1986. After that he missed a total of 6 games in 10 full seasons with the Bulls.

Your 10 year-set is taking 2.2 games per season from Russell and then throwing less than a game a season for Jordan alongside a larger sample from one season that does not compare favorably with Lebron, Kareem, Hakeem, Bill, or various other players I've neglected to mention(KG, Shaq, ect). And for all that, if we account for certain eras requiring lower SRS for high championship probability...

At the height of their dynasty, the Celtics were comically dominant. From 1962-65, their average margin-of-victory (MOV) was over 8 points per game. During the same time span, only two other teams even eclipsed 4 points per game – the ’64 Royals and the ’64 Warriors. And all of Boston’s separation was created by its historic defense, anchored by Russell:

...Jordan is still well behind Russell(and by extension Wilt):

Notably, if we take WOWYR seriously, Bill Russell led the greatest team ever with 35 win help throughout his prime while Jordan barely won half as much with 40-50 win help. While Jordan looks marginally better than Lebron, he's not really within range of GOAThood.

Just to give Jordan the slightest empirical advantage over Lebron, we reduced our per-season sample by a factor of ten and Jordan still comes out well behind Russell and Wilt(as well as Magic and D-Rob).

Regressed or non-regressed, if we use larger samples, Jordan plummets relative to Lebron and Kareem. We can discuss the merits of this type of data, but trying to paint his WOWYR stuff as some GOAT-lvl indicator doesn't really work. Notably, the disparity generally comes from the defensive side of things. A disparity we shouldn't be surprised box-stuff can't account for. It's also a disparity many of the theoretical excuses made for the lack of comparable influence doesn't really consider(The 2015 Cavaliers say hello).

Lots of interesting thoughts here! Let me try to organize them into a few main points, but do let me know if I missed any!

Claim 1: “Regressed or non-regressed, if we use larger samples, Jordan plummets relative to Lebron and Kareem. [and Russell, etc.]”.
So right off the bat, this is not true. WOWYR looks at 10-year samples, and Jordan comes out ahead of LeBron, Kareem, Russell, and all the other players you mentioned. In GPM, again using a 10-year sample, only Russell comes out ahead, while Jordan comes out ahead of LeBron, Kareem, Wilt, Hakeem, etc.

You provide lots of interesting raw data! And for the record, I do find the raw data quite compelling for people, Russell in particular since our stats for him are so limited. There’s a reason I have Russell in my GOAT tier!

But remember, these stats are “UnAdjusted”…. and unadjusted plus minus data can make 2001 Playoff Derrick Fisher look better than 2016 Playoff LeBron. They make no corrections for opponent (they use Margin of Victory, not SRS), nor teammate health, nor opposing health, nor overall team context. So if the Adjusted data disagrees with the raw data, I’m personally more inclined to believe the Adjusted data, or at least heavily consider its validity.

Claim 2: “WOWYR is based on small individual sample sizes, unlike single samples of raw WOWY in longer periods where players miss time”.
For example, you cite that WOWYR only uses 2.2 games per season to count Russell’s “off” sample, and imply there’s a similar minuscule number of games to calculate Jordan’s off sample. I’m not sure this is true.

Remember, WOWYR is Adjusted for teammate context, and so it uses the WOWY of every other major player on the team to help isolate Jordan’s. The graphic in the WOWYR article I linked above shows this.

In other words: Jordan’s numbers aren’t just based on the 666 games he played and the 154 games he missed in the 10-seasons we’re looking at (154 -> 72 games out if we ignore 94). We can also use the games when other players were in or out of the lineup: Pippen was out for 113 games in this timespan, Rodman was out 18 games, Horace Grant was out 28 games, Bill Cartwright was out 66 games, Luc Longley was out 102 games, John Paxson was out 86 games, early guard Sam Vincent was out 81 games, early forward Gene Banks was out 19 games, early center Earl Cureton was out 39 games [at least by my count]

… every one of these games with lineup changes goes into the Adjusted calculation for WOWYR. And every one of them is used to tell us a little bit more about Jordan’s specific value. Now it’s not as much as we would need to decrease the WOWYR noise to as low as we’d get for RAPM, but that seems like a lot of games to me!

Regardless, it’s certainly a heck of a lot more than just using a handful of games a season to find Jordan’s raw WOWY. This kind of increased sample size would also apply when calculating WOWYR for players like Russell, Kareem, LeBron, etc. That’s one of the benefits of WOWYR over raw WOWY… the fact that we’re adjusting for teammate-context allows us to use the teammates’ games in and out of the lineups to increase our sample size pretty dramatically.

(2) The effect of good coach on WOWY: WOWY is sensitive to coaching. If a player is missing one game, for example, a poor coach may put less effort into adjusting the game plan than if a player is missing a lot of time in a row. If one coaching staff is much better than another coaching staff, the better one might do a much better job at filling in when a player is missing than a bad one. Jordan had great coaching with Phil Jackson. This is the kind of thing that would limit his raw WOWY.

The assumption that good coaching must depress impact is questionable. In fact I would say bad-coaches failing to optimize a star player's influence is as common than the opposite. I'm also not sure why you're using a one-game sample in your analogy when its multiple 82 game samples that are being relied upon the most here. If your theory holds, then the relatively pedestrian stuff we have under Jackson's significantly worse predecessors should be sparkling. Moreover, if you're concerned about coaching ajustments, then WOWY is really the way to go here, as it's much easier to make adjustments with some pre-time before a game or a season, than it is to make adjustments when a player leaves half-way through.

Frankly this is an exceptionally weak approach to take with say, Lebron, considering that Lebron looks better in larger samples of "off", and his teams tend to look the worst when the team is given time to adjust. This includes his time under Erik Spoelstra where the heatles without Lebron did not look as good as the Bulls without Jordan or the Bulls without Jordan and Grant(at least by SRS).

Not to be too critical here, but this seems like another example where you've come up with a seemingly viable theory, without actually looking where the breadcrumbs lead.

WOWY(and WOWYR) is indeed noisy, which is why it's good to look for replication across a variety of contexts:

somewhat behind the best stuff we see from Hakeem(25 and 30 game lift in 20 game samples in 88 and 90), consistently behind Kareem throughout the 70's(30 win lift in 75, a 29 win improvement with a player similar to oakley as a rookie, 62 wins without his co-star, and takes the depleted remnants of a 30 win team to 45 wins in 77), and a pretty sizable gap compared to Lebron who has multiple 40 win signals for 09 and 10, 30 win signals in his second cavs stint, and is mostly operating at, at least 20+ win lift throughout his prime leading multiple teams to 60 or near-60 win basketball without co-stars on top-heavy rosters(cavs, heatles).

The disparity is consistent. I don't need to cherrypick one year or approach to observe a gap. That's a pretty good indication that this can't just be put down to "noise". It's good to look at everything and assess the evidence holistically. The best possible signal I can get for Jordan is to take record instead of srs for 1986, ignore 84, **** with the minute thresholds ,ignore Ben's much more pedestrian appraisal, pretend Oakley didn't help them defensively, and you get 32 win-lift for a sub 50 win team. That took many, many extra steps and it still doesn't get you to what Lebron does in 2009, 2010, or 2015 and 2016. That 23 win-appraisal I throw around works on the assumption there was no improvement after MJ was drafted(again, ignoring evidence that Oakley helped alot defensively). "Impact" is just not a winning case unless you ignore the forest for trees.

Re: coaching, I’m not married to the theory. But let’s return to 2001 playoff Derrick Fisher vs 2016 Playoff LeBron. Raw plus minus says Derrick Fisher is better than LeBron. However, when we adjust for teammates and context, we come to the (obvious conclusion that) playoff LeBron is actually better. If we’re tight on time, we can trust that the math works out that way and move on, or if we’d like to dive a little deeper, we can start to contextualize what made Derrick Fisher’s raw numbers mislead us. The context I’d personally apply is the fact that his minutes aligned with prime Shaq, Kobe, and a variety of other talented teammates.

Now we return to WOWY and Jordan. Raw wowy suggests Jordan is worse than some of his all-time competitors. But when we adjust for the context (exactly like we did above), we find that Jordan is suddenly more valuable than LeBron and Kareem and many other all-time players in their 10 year span. The coaching was just my attempt to explain why Jordan’s raw data might underrate him, just like I attempted to propose context for why Derrick Fisher’s raw data overrated him. You might believe the argument, you might not, you might have other ideas (I’m all ears!)… but these contextual arguments don’t change the fact that, mathematically, when we Adjust Jordan’s raw WOWY data, he ends up on top over the other GOAT candidates.

Now you might make arguments for why those other players are still better than Jordan. You might cite the nosiness of WOWY-based data. Your comment that margin of victory might underrate Russell given the small league is an interesting one! But to return to the premise of the thread: WOWYR is another impact stat that portrays Jordan as GOAT level, over Russell, Lebron, and Kareem.

D. RAPM: Jordan, LeBron

Speaking of which...

We have less than 20% of Jordan's games measured in Jordan's 6 best years, we have strong evidence that the games we do have underrate Jordan, and Jordan nonetheless comes out 8th all time, tied with peak 2013 LeBron. Again, LeBron's non-peak years may still end up being higher than Jordan.

As far as the data you're actually using is, 2013 Lebron is not Lebron's peak, and the comparison here is Jordan vs Lebron, not "Jordan vs Miami Lebron" or "Jordan vs whatever year of Lebron might give MJ a semblance of a case". You are more than smart enough to recognize the difference between letting the evidence speak and strangling it so that it fits your priors.

You also neglect to mention that the data we have for 1988 actually skews in Jordan's favor as the Bulls did worse during the portion of the season we don't have data for. As I'm sure you're aware, there's plenty pointing to 1988 as Jordan's most "imapctful" season and that conclusion would actually fit the "bad coaching" theory you offered earlier. It is fair to point out uncertainty, but trying to take data that clearly favors Lebron as actually "pro-mj" because more data may improve how Jordan looks is a bit of a leap. As it is, we do have playoff on/off here, and it doesn't support that conclusion. Jordan's on/off arcs downward(in line with a defensive decline observed in both Blocked and Ben's film-tracking) from 1988 to 1993 before rebounding for the second-three peat. As it is, Jordan's 1988 also scores near the top in the offense-skewed stuff you seem to prefer, so this honestly seems like a questionable prediction.

Also important to note, before we use "tiers" to explain this away, this data only really exists for the peaks of post 1997 greats(and MJ), so Jordan only looking sizably worse than one modern player(doesn't really look better to me than duncan or kg here though maybe an expert like Jaivi can offer some distinction), doesn't mean he's "close" to being "the greatest". He flatly doesn't score close to Lebron here, and we have no way to know if that would apply to players like Wilt, Russell, or Kareem. Notably, RAPM consistently places Lebron well ahead of the likes of KG, Shaq, and Duncan(I recall seeing a 5 year average where the gap between lebron and 2nd place KG was similar to the gap between KG and 7th place Nash), players who, with more "apple to apple" pure impact analysis look quite comparable to Jordan. Crude comparison, but at this point it's a straw on a camel that broke yesterday. It's not as emphatic as WOWY(regularization will do that), but it's just not a winning case for puffy-j.

Good points here. A few thoughts.

Re: “2013 LeBron is not LeBron’s peak”, most people consider LeBron to have peaked in 2013. You can see in any past Greatest Peaks project on this board and find 2013 LeBron voted in over other years. Look on other sites and other talk shows and people will normally say 2013 LeBron. This brings us to the question of why RAPM rates 2013 Lebron lower than other years, or whether it’s missing anything. Perhaps the larger group thinks there’s a greater playoff resilience in 2013 than in 2009? Or perhaps the larger group is just mistaken? One possible explanation for why 2013 is lower is that RAPM measures impact in a specific role… perhaps LeBron’s role in 2013 was less suited to maximize his individual value. (You’ll note that this is the exact argument made against LeBron by scalability proponents).

Re: 91 Bulls vs 88 Bulls: The 1988 Bulls over performed by 1 single game: they were on a 51-win pace in our sample and they won 50 games in the full season. The 1991 Bulls underperformed by at least 8-9 games: they were on a 52/53-win pace in our sample and they won 61 games in the full season. They were on a 64-win pace by SRS. 1988 is well within random variance; 1991 Bulls is the largest underrated sample in the entire Squared2020 database.

You can argue that RAPM still favors LeBron. Personally, I’m open to the idea. But since the thread asked “are there any impact metrics that can argue Jordan is GOAT-tier”, it seems reasonable to say that a Top-8-all-time number in a sample that drastically underrates how good the 91 Bulls were and doesn’t measure Jordan’s value in his other peak years of 89 or 90, or other near-peak years of 92 or 93 might suggest GOAT-tier or near-GOAT-tier impact.

Re: definition of GOAT tiers, see my first comments above In short, perhaps we just have different definitions of what GOAT-tier is. Which is okay!

E. Is Duncan GOAT-tier??

To me, a more fair characterization of the box-approximations of PIPM and RAPTOR is that they predict true plus-minus-based PIPM and RAPTOR, with wider error bars than the plus-minus based ones.

Sure, but it's not just "wider error bars", it's "wider error bars largely because they ascribe outsized(relative to historical precedent and actual impact signals) defensive value to smaller steal and block accumulators." But even then, looking at the metric that accounts for defense best...

And that if Jordan looks comparable to LeBron

But he doesn't. You specifically chose a favorable frame of comparison for Jordan(3-years consecutive), and Lebron has, not one, but two better stretches when we utilize that frame. Going off the data RK listed, Lebron has the 2 highest scoring years(with 2009 being far ahead of anything else), and 5 of the best scoring 7. I could literally chuck the best scoring year by far, and Lebron would still look better. Jordan does not look comparable, and he does not rank 3rd-all time, he ranks 3rd among the players we actually have data for. PIPM dates back to 1977. That leaves at least 2 players with consistently better impact indicators completely out of the room.

Agree to disagree here. I’ve provided plenty of evidence for why I they’re comparable statistically, and you’ve provided plenty of reasoning against it. It’s okay to not always agree with everyone!

As for AUPM, you can shake off Lebron if you use a three-year frame(note I said "generally speaking" and "most comparative frames" as qualifiers), he still falls short here to Duncan. Considering that AUPM is constructed as a combination of on/off and BPM, that Duncan grades out #1 here is rather impressive, especially since we are using a frame of comparison(3 years consecutive) that gives him the best looking case. And remember, Jordan does not score "2nd all-time", he scores "2nd among a minority of historical players in this specific metric using the most favorable possible comparative lens". Considering you don't have Duncan to have "GOAT tier" impact indicators, it seems logically inconsistent to me to argue that impact-data potrays Jordan as a "definite GOAT-tier player"

...Duncan scores higher in aupm despite aupm being partially constructed with BPM, scores as high as a pretty optimistic MJ WOWY appraisal in injury plagued 04/05, looks similarly dominant in RAPM stuff(though this gets very noisy with different scales), and won 57 and 62 wins at his most valuable looking stretch as opposed to 50 for Jordan in 1988.

Hakeem looks better if you use his very best WOWY samples, looks better in his first three years, and looks similarly impactful throughout his prime, while scoring higher in postseason PIPM(the box metric which most closely is tied to actual defensive impact.(remember that pre 97, none of the "plus-minus" stats you reference have plus minus(or film tracking)). Hakeem also scores similarly in 97/98 on/off despite arguably being further from his peak than Jordan was those years.

I don't mind different definitions, but I think its a good idea to keep our thresholds consistent. When you tell me someone has GOAT-tier impact stuff, I want to see something that suggests you were the greatest. Maybe you're using a more liberal definition, but I don't think consistent application leaves Jordan significantly above TD and The Dream.

I think this is flat-out untrue. Jordan's playoff Augmented Plus Minus, based on actual plus minus data, is better than LeBron's. His WOWYR is over LeBron/Russell/Kareem. And all the approximations of more accurate stats we have show him as GOAT tier.

In AUPM, Jordan looks worse than Duncan and better than Lebron in one framework while looking worse than LBJ in most others. That is also just a fraction of nba history being accounted for.

His WOWYR is flatly worse than Russell's(and Wilt) over 10 years(when we adjust for lower srs-championship tresholds), and when we take >10 or 82 game samples instead of a sample of 6 games over 8 years, Jordan scores well behind whether you prefer WOWY or "corrected WOWYR. I also don't know what you're basing these metrics being "the most accurate" from. The box-stuff specifically gets less noisy by chucking out defensive accuracy and the more useful method(imo) where we just replace the defensive box-score with defensive impact, immediately sees MJ plummet.(Jordan is tied or ahead of Lebron in D-RAPTOR, ahead of Kareem in BBR D-BPM, well, well behind on D by basically all impact stuff).

Perhaps these stats aren't as bad as PER, but nonetheless, they skews heavily towards Jordan(at least relative to the history of great defenses, and the "real" impact signals of the players in question), and Jordan still does not get #1 if it has actual on/off or plus-minus. Coincidentally, his actual on/off looks much, much worse, as does WOWY and adjusted WOWY over serious samples(>10 games).


If you loosen you definition of Impact(non-plus minus RAPTOR and pure Box with weak correlates) you can get Jordan there(along with someone like Duncan), but I feel my definition of "impact" is more in spirit with what impact denotes and is ultimately more useful.

Accept or reject that, but consistency is key:

...Duncan scores higher in aupm despite aupm being partially constructed with BPM, scores as high as a pretty optimistic MJ WOWY appraisal in injury plagued 04/05, looks similarly dominant in RAPM stuff(though this gets very noisy with different scales), and won 57 and 62 wins at his most valuable looking stretch as opposed to 50 for Jordan in 1988.

Hakeem looks better if you use his very best WOWY samples, looks better in his first three years, and looks similarly impactful throughout his prime, while scoring higher in postseason PIPM(the box metric which most closely is tied to actual defensive impact.(remember that pre 97, none of the "plus-minus" stats you reference have plus minus(or film tracking)). Hakeem also scores similarly in 97/98 on/off despite arguably being further from his peak than Jordan was those years.


IIRC, you have dismissed both Duncan and Hakeem as having GOAT-level data on multiple occasions. If Jordan's impact stats potray him as "absolutely GOAT-Level at his best", why don't you extend that for Hakeem and Duncan who do just as well if not better using data which actually has "impact" in it.

Is Duncan a GOAT candidate according to "impact"? If so, sure, put Jordan there. If not, then I don't think MJ really has a case(at least if "impact" is the lens).

And yes this post was brought to you by the San Antonio Spurs

Love the last comment!

I'm open to suggestions that Duncan's peak is GOAT tier, at least statistically. I personally have Duncan a touch below ~5th GOAT peak, but he does have a good high-end argument. Duncan's also great on the longevity front. My concern for Duncan is the 'width' of his peak and his overall prime-ability being a touch below the other GOAT-tier players.

If I'm trying to measure a player statistically, I like to look at a variety of stats. Duncan's plus-minus based stats are indeed GOAT level.But Duncan grades a bit lower in box-based stats and WOWY-based stats (still great all-time, but under all the GOAT-tier players we've been discussing).

I've pretty thoroughly discussed Hakeem's lower performance in the statistics in other threads, not looking to get too far down that rabbit hole here.

[MyUniBroDavis]

^ that 2013 is voted as brons peak does not mean that it is. I think ohayo has it either as 09 or 2016

When we talk about Jordan as a GOAT candidate I’m curious if we can differentiate his playoff impact from his RS impact since he’s a playoff rider

For what it’s worth, in terms of relative to competitions def rtg, there are issues with doing it this way but it’s fun and prolly better than the raw values

Postseason offense vs opponent defense (not weighting longer series more)

1991 bulls were +11.5 on offense
1992 bulls were +7.7 on offense
1993 bulls were +10.05 on offense

The magic Lakers from 85-88

1985 +10.9
1986 +7.6
1987 +10.9
1988 +6.8

I’m curious where they rank compared to other playoff title or finals offenses. It’s lower than the 2016 or 2017 cavs and I think the 2017 Warriors off a quick check. Might be fourth

There aren’t any real impact metrics I feel that are worth anything throughout his prime.

When we’re evaluating the top offensive guys I think it’s fair to say that Jordan is in the discussion for the GOAT offensive player, and pretty much everyone that’s also in that discussion is either a neutral or negative defender outside of Bron so it would be probably based on that.

[f4p]

^ that 2013 is voted as brons peak does not mean that it is. I think ohayo has it either as 09 or 2016

When we talk about Jordan as a GOAT candidate I’m curious if we can differentiate his playoff impact from his RS impact since he’s a playoff rider

For what it’s worth, in terms of relative to competitions def rtg, there are issues with doing it this way but it’s fun and prolly better than the raw values

Postseason offense vs opponent defense (not weighting longer series more)

1991 bulls were +11.5 on offense
1992 bulls were +7.7 on offense
1993 bulls were +10.05 on offense

The magic Lakers from 85-88

1985 +10.9
1986 +7.6
1987 +10.9
1988 +6.8

I’m curious where they rank compared to other playoff title or finals offenses. It’s lower than the 2016 or 2017 cavs and I think the 2017 Warriors off a quick check. Might be fourth

for 2017 warriors, the real number would probably be +9.5. it's +11.7 overall, but that includes an obviously affected +18.7 from the kawhi injury against the spurs, a result that obviously wasn't going to happen pre-injury.

based on that 1997 jordan thread, i'm starting to think i've slightly overrated him defensively but underrated him offensively. scottie pippen doesn't seem like a #2 offensive weapon good enough to explain the offensive results of the bulls.

[MyUniBroDavis]

^ that 2013 is voted as brons peak does not mean that it is. I think ohayo has it either as 09 or 2016

When we talk about Jordan as a GOAT candidate I’m curious if we can differentiate his playoff impact from his RS impact since he’s a playoff rider

For what it’s worth, in terms of relative to competitions def rtg, there are issues with doing it this way but it’s fun and prolly better than the raw values

Postseason offense vs opponent defense (not weighting longer series more)

1991 bulls were +11.5 on offense
1992 bulls were +7.7 on offense
1993 bulls were +10.05 on offense

The magic Lakers from 85-88

1985 +10.9
1986 +7.6
1987 +10.9
1988 +6.8

I’m curious where they rank compared to other playoff title or finals offenses. It’s lower than the 2016 or 2017 cavs and I think the 2017 Warriors off a quick check. Might be fourth

for 2017 warriors, the real number would probably be +9.5. it's +11.7 overall, but that includes an obviously affected +18.7 from the kawhi injury against the spurs, a result that obviously wasn't going to happen pre-injury.

based on that 1997 jordan thread, i'm starting to think i've slightly overrated him defensively but underrated him offensively. scottie pippen doesn't seem like a #2 offensive weapon good enough to explain the offensive results of the bulls.

Tbf another thing to take into account is the Warriors rested a bunch of fourths and we’re meming around lol

I feel it’s not too suprising though, Jordan’s a GOAT peak contender, most people have him there although I have 16-18 bron ahead by a pretty good margin, his offense being the main driving force means this isn’t too suprising I think though

[OhayoKD]

Well, again we go.

A. Intro and ‘What is GOAT tier’?

You’ve suggested Top 3 in a post-1997 stat isn’t high enough to be considered GOAT-level.

Here’s the problem: there is no absolute-consensus #1 player across all impact metrics, not since 2014, not since 1997, not since 1977, not since 1955, not since 1946, not in peak, prime, or career. There just isn’t.

Ah, but "absolute consensus #1 player across all impact metrics" is not the standard I'm holding Jordan to. It's more like #1(among greats) in at least one metric. And, of the stats we've discussed as "impact"(using a wider umbrella than I'd prefer), the best Jordan manages, if we use his best comparative frame, is #2 behind Duncan. In general(meaning a majority of potential comparative lenses) he also scores behind Lebron. Even with most of nba history completely cut-out, Jordan doesn't meet my bar.

This is also an especially weird point to make in a comparison with Lebron James who pretty much reaches the standard of "absolute-consensus #1 player across all impact metrics" for the data-ball era. Most RAPM-Data sets have Lebron miles ahead of anyone post-1997. Using Ben Taylor's own 5-year RAPM, the gap between Lebron and 2nd place KG is as big as the gap between KG and 7th place Steve Nash. Looking at his prime as a whole...

2018 is the 25th season of league-wide plus-minus data, which covers nearly 40 percent of the shot-clock era and touches 12 of the top-20 players on this list. None have achieved LeBron’s heights: He holds four of the top-five scaled APM seasons on record, and six of the top eight. Since 2007, 10 of his 11 years land in the 99th percentile.

...that seems pretty close to a "consensus #1" to me.

Lebron has the best scores in squaredcircle's data, puts up the best 7 scores(minuites filter is at play) using Owly's data, and is pretty much smoking a set of players who compare pretty favorably to Jordan if we use "real" signals for an apples to apples comparison.

Lebron dominates PIPM, has a WOWY/WOWYR profile only really rivalled by Kareem, Russell and Wilt(more on that later), looks the best in most AUPM comparisons, looks the best in Circle's "playoff on/off" sample(role player Robinson notwithstanding). and is 1 or 2(depending on comparative frame) in pretty much every box-metric(becoming a clear 1 if you so much as hedge between defensive impact and defensive box-score).

In short, Lebron does very well, even with this new standard you've created, while Jordan is struggling to meet a much lower bar(in-era impact stuff also sees him directly challenged by Hakeem and Magic). Lebron's only real competition post-97 comes with Curry who is competitive specifically with 30+ Lebron(in some metrics) in the regular season, before falling well short in the playoffs. If Lebron's "impact metrics" are the bar for GOAT-Tier, then Jordan just isn't there.

Using "rankings' like "top 2" or "top 3" may give the impression of parity, but with the exception of AUPM, the margin is sizable, and this naturally compounds when you greatly extend the amount of players counted(which is exactly what happens when we use "raw signals" where the data for historical comparisons is available). Speaking of which...

So right off the bat, this is not true. WOWYR looks at 10-year samples

You seem to be confusing number of years with number of games. Critically, if you are looking at a similar number of games over a larger time span, your per season sample becomes smaller and all those extra years can make the data less accurate. Moreover, this is just flat-out false:

Regardless, it’s certainly a heck of a lot more than just using a handful of games a season to find Jordan’s raw WOWY.

No. The "Anti-MJ" data consists of 82 games in 1994, 82 games in 1984, 62 games in 1986, and an additional half-season in 1995. His disadvantage stays whether you use regressed data, non-regressed data, srs, or record. As I just illustrated, using regressed data from 1986, 23 year old MJ ends up looking less impactful than Lebron at the ages of 19 and 20. The samples here I'm working off are much bigger than what you're using, per-season and overall. To claim this is all based on a handful of games per season is just wrong.

Otoh, that 72 game you're working with comes out to 7.2 games a season. You can apply as many teammate adjustments as you want, but there's very little to apply them to. Moreover, most of those 72 missed games come from 1986, which as we've covered, taken in isolation, as opposed to being presented as indicative of Jordan's cast throughout his prime, does not paint Jordan favorably. So really, your 10-year WOWYR estimation for prime MJ is basically just based on a season a game.

Honestly, looking into these adjustments, I think this gets worse:

Pippen was out for 113 games in this timespan, Rodman was out 18 games, Horace Grant was out 28 games, Bill Cartwright was out 66 games, Luc Longley was out 102 games, John Paxson was out 86 games, early guard Sam Vincent was out 81 games, early forward Gene Banks was out 19 games, early center Earl Cureton was out 39 games

For the sample I think you're using(1987-1997), Pippen only missed significant time(as a starter) in 1989(30ish games) and 1994(10). IOW. this "10 year adjustment" is mostly based on Pippen's exploits as a second-year player.. The "adjustment" being applied to Jordan's ben teammate isn't "correcting" the data, it's distorting it. Not only have we reduced our sample size by a factor of 10, but we're also treating Pippen's exploits as a second-year player as if they are relevant to what Pippen was doing in 1991. (Side-bar: You seem to be including games where rookie Pippen didn't start from 1988 while excluding games where Grant sat in your "without" here. Is there a reason for this?)

This is actually an issue Ben himself outlines(check those articles you linked), and probably why in his own impact evaluations, he mostly focuses on concentrated stretches, not "10-year samples" featuring about a 10th of the relevant evidence. Moreover. even after we've used a 10-year regression to turn a mountain of data into pebbles, Jordan is still well off the very best:

Your comment that margin of victory might underrate Russell given the small league is an interesting one!

You say "interesting", but I'd say its "essential". Unless you're more interested in regular season win totals than championships(or championship probability), saying that WOWYR puts Jordan over Russell(and by extension, Wilt) is just misleading. WOWYR says Russell won 11 rings with 35-win help before adding that Wilt was a relative peer. Jordan isn't there. Use large samples, and Jordan looks well off the big-three(Lebron, Bill, and Kareem) while also getting hounded by Duncan, KG, Hakeem, Walton, Shaq and so on. It's not GOAT lvl data, at least not by the standard we seem to be using with Duncan or Hakeem.

But remember, these stats are “UnAdjusted”…. and unadjusted plus minus data can make 2001 Playoff Derrick Fisher look better than 2016 Playoff LeBron. They make no corrections for opponent (they use Margin of Victory, not SRS), nor teammate health, nor opposing health, nor overall team context.

Well one, most of the "anti-mj" stuff we've been using uses SRS, not "record". In fact, Jordan generally looks worse when you use srs to assess teams as opposed to a team's record. More importantly, in the specific cases of Lebron and Jabbar, we're not dealing with a "one-off" involving a small sample of games...

somewhat behind the best stuff we see from Hakeem(25 and 30 game lift in 20 game samples in 88 and 90), consistently behind Kareem throughout the 70's(30 win lift in 75, a 29 win improvement with a player similar to oakley as a rookie, 62 wins without his co-star, and takes the depleted remnants of a 30 win team to 45 wins in 77), and a pretty sizable gap compared to Lebron who has multiple 40 win signals for 09 and 10, 30 win signals in his second cavs stint, and is mostly operating at, at least 20+ win lift throughout his prime leading multiple teams to 60 or near-60 win basketball without co-stars on top-heavy rosters(cavs, heatles).

If this was just us looking at the Celtic's one-year turnaround with rookie, I'd be more sympathetic, but as has been covered, Lebron and Kareem's advantage holds throughout their careers at multiple spots in a wide variety of contexts. The gap peaks with the largest possible samples and maintains even in situations where "scalability" theocraticals predict it shouldn't(2015, 2020, 2012, 2005 and 2006). You are not dealing with a few games of Derick fisher, you are dealing with a mountain of data extrapolated from various methods and saying it's comparable (or worse) to grains of sand. If you're that concerned about teammates, why not just look at roster-changes, schemes and adjust as opposed to trying to correct 1991 Jordan's off with a bunch of games in 88without Rookie Scottie.

(1) One could make some sort of team-building relative-to-position argument in favor of Jordan... i.e. that Jordan's a bigger defensive outlier among guards than Kareem is among bigs,

Sure.

(2) I'm also open philosophically to the idea that Jordan's all-time motor leads him to 'get the most' out of his defensive value more consistently as he gets older.

Well, again, "impact" and team signals aren't really on your side. 2015-2017 Lebron looks more impactful than any Jordan defensively in the regular season and the gap widens in the postseason(Not only do the cavs elevate against playoff opponents, they elevate against top 5 offenses. Whether you go by DRAPM, ON/OFF WOWY, net-rating, ect, Lebron consistently grades out as a more valuable defender for the regular season, let alone the playoffs where a 13-year sample(not even including 2020) looks comparable to a comparatively sample from Kawhi(averages go down the longer you play). You really have to look at the very worst defensive years(at least emperically) for a non-box comparison to become favorable(and they aren't worse than the worst stuff we see from Jordan). I'm not sure how it plays out with Kareem, but I imagine a similar story unfolds given Kareem's able to mantain his "holistic" impact edge throughout the 70's.

(1) Even if steals are overrated in box-based defensive measurements (e.g. if the box stats miss the fact that the steals come expense of unnecessary gambling, as they may for Jordan), steals are still individually the most valuable defensive play someone can make.

So a couple notes here.
(1) You are combining the defensive value of a steal with the offensive value generated. On the defensive side alone, steals aren't nearly as valuable as plays at the rim. Additionally, just like blocks, "steals" from a non-big often are a byproduct of a bigger player's influence...
https://youtu.be/p5aNUS762wM?t=1165
Here, Jordan is able to get a steal because Oakley stonewalls the attacker and occupies his attention. Yet as far as these box-models are concerned, all the credit here belongs to MJ. Notably, it was Oakley's arrival that saw the Bulls become a -2 defense in 1988(the only good defense Jordan has ever anchored), and it was with Oakley's depature that the Bulls fell back to mediocrity. Charles did not rack up enough steals or blocks for stuff like "RAPTOR' to love him, but I'd argue on plays like these, its Oakley who deserves most of the credit, not steal-getter MJ.

(2) Whatever the value of a steal, these metrics have no way to account for when steal attempts fail. Considering Jordan consistently posted high-error rates(dropping to the 14th percentile in 1991 despite a drop in defensive activity), only including the positives and completely excluding the negatives will naturally inflate how good a guard looks. We can actually see this if we compare Jordan's steal-d-rating correlation with Kawhi's:

Kawhi doesn't rack up as many, but because he gambles significantly less, his steals end up having a stronger influence on the quality of his team's defense.

Unless these sorts of things are somehow accounted for in these box-components,(as far as I know BBR BPM actively compounds the issue by giving smaller players more credit for blocks) I'm pretty confident high-gambling steal-getters(and to a degree, undersized block accumulators) are getting significantly juiced. However, when possible it's better to see if we can back our theories with evidence as opposed to speculation, which is probably why, contrary to whatever "consensus" you are referencing, Ben Taylor specifically argues that MJ's playoff on/off is important so we have a way to properly account for his defensive impact:
https://youtu.be/p5aNUS762wM?t=1291. And yes, on/off is "noisy", but if you're really concerned about noise...

(2) Some of the better box-stats I'm using have input weights that are calibrated overall, rather than calibrated for offense and defense separately. In other words, we calculate those stats as a whole, then split into defensive and offensive components after. This may effect the error: in this case, a player that’s overrated on defense may actually be underrated on offense (since the mistake is whether we attribute the value to offense or defense, not in the overall value). Not a guarantee, just a possibility. Like you say, it may also mean they’re overrated overall.

...it's probably a good idea to reference larger samples. Maybe there is an allocation problem here, but the largest possible samples(including multiple 82 game-sets), actually have the defensive gulf being bigger than what one might extrapolate from on/off. This tracks with the Bulls going from average to #1 with Jordan's defensive activity and effiency(breakdowns/error rate) declining, them staying average defensively until Oakley's addition, his relatively unimpressive D-RAPM, the Bulls effectively being unaffected on defense by Jordan's departure, and even his D-PIPM(IIRC Ben says historic PIPM does a better job accounting for defense than similar metrics, something to do with using linear-regression as opposed to tree-branching?) which, outside of 1988, looks worse than basically all of Lebron's prime save for a couple down-years(Lebron is able to match 1988 at various points fwiw).

All considered, the real world seems to disagree with the box-one on defense with this type of archetype, and it does so consistently with the disparity not really getting any better if we look at raw individual data or even the history of great defenses.

Re: “2013 LeBron is not LeBron’s peak”, most people consider LeBron to have peaked in 2013

But since when was "consensus" good justification to throw out the vast majority of data as "noise"? The data you are using disagrees. Even if you disagree with what the most straight-forward interpretation here, taking the worst signal to argue they're "comparable" and ignoring the vast majority of data giving a single player an edge is bad, bad practice. As Ben would say, we shouldn't just look at players at their lowest statistical points, or their highest statistical points, we should look at things holistically. Even if you're right and 2013 is Lebron's peak, the majority of the data putting Lebron signficantly ahead(with the worst looking singal merely looking "as good") is a clear win for Lebron. This can be applied to PIPM too. And if you're going to dismiss data that hurts Jordan because "it's not lebron's peak!", appealing to consensus really isn't enough here. As it is, I don't think I've seen you really address the abudance of examples where Lebron is able to generate jordan-level or jordan+ data, in theoretically sub-optimal conditions(or at least what "scalability" predicts would be sub-optimal).

But that's really besides the point here. If a player's worst set of data(this applies to PIPM too) still looks comparable to another player's best, painting it as anything but an empirical win is just disingenuous.

Point taken on 88, but again, 88 and 91 are the two years that score the best for MJ by a variety of metrics(and with defense accounted for we see a steady decline in on/off despite of a box increase between 88 and 91). Considering Lebron has several years, scoring well ahead of either data point, this'll only take you so far. Frankly, I'm not even sure why you're confident these excluded years would be significantly better than 1996. The Bulls won 70 games and Jordan's raw stuff puts him at 20ish wins and he and the Bulls benefit(like Kareem does to a degree) from the perks of expansion.

Additionally, with RAPM, we get into regularization where outliers get scaled down at around 25-30 win lift(and Lebron, unlike Jordan is hitting or breaking those marks several times, in various contexts, over sizable samples).

Regardless, if scoring 8th in a narrowed field puts you at the top-tier, I think we need to start being more generous with who we consider viable candidates here.

I’ve provided plenty of evidence for why I they’re comparable statistically, and you’ve provided plenty of reasoning against it. It’s okay to not always agree with everyone!

Maybe I missed something? All I recall is you highlighting that Jordan's best 3-year PIPM stretch is a little ahead of Miami-Lebron(while being behind two other sets). I think I addressed that above, but if there's other evidence I skipped, feel free to point it out!

Re: definition of GOAT tiers, see my first comments above In short, perhaps we just have different definitions of what GOAT-tier is. Which is okay!

Sure.

I'm open to suggestions that Duncan's peak is GOAT tier, at least statistically. I personally have Duncan a touch below ~5th GOAT peak, but he does have a good high-end argument. Duncan's also great on the longevity front. My concern for Duncan is the 'width' of his peak and his overall prime-ability being a touch below the other GOAT-tier players.

If I'm trying to measure a player statistically, I like to look at a variety of stats. Duncan's plus-minus based stats are indeed GOAT level.But Duncan grades a bit lower in box-based stats and WOWY-based stats (still great all-time, but under all the GOAT-tier players we've been discussing).

Well, Duncan does score #1 in AUPM(despite it really being something that should favor MJ), he looks as impressive to me in RAPM(though again, this gets tricky with different creators/scales), compares well in rapm to players like Shaq and KG(who do well relative to Jordan in apple to apple comparisons), and looks a bit better(if we use SRS for MJ instead of record or account for Oakley) in a couple of injury plagued seasons using the raw stuff(srs or record I think). Say what you want about the noise, but that is a 15 game/season sample(larger than anything you've brought up I think).

If we open the floor to box stuff, then yeah, MJ gains an advantage(as we would expect given what was discussed earlier), but a pure impact comparison looks favorable. Not going to speak too much regarding the quality of his extended prime, but its worth noting that if you place their best couple of years as even comparable(and there's plenty of reason to), "CORP" outputs flip to giving Duncan an edge.

And yes, this too was paid for by Gregg Popavich

[falcolombardi]

^ that 2013 is voted as brons peak does not mean that it is. I think ohayo has it either as 09 or 2016

When we talk about Jordan as a GOAT candidate I’m curious if we can differentiate his playoff impact from his RS impact since he’s a playoff rider

For what it’s worth, in terms of relative to competitions def rtg, there are issues with doing it this way but it’s fun and prolly better than the raw values

Postseason offense vs opponent defense (not weighting longer series more)

1991 bulls were +11.5 on offense
1992 bulls were +7.7 on offense
1993 bulls were +10.05 on offense

The magic Lakers from 85-88

1985 +10.9
1986 +7.6
1987 +10.9
1988 +6.8

I’m curious where they rank compared to other playoff title or finals offenses. It’s lower than the 2016 or 2017 cavs and I think the 2017 Warriors off a quick check. Might be fourth

for 2017 warriors, the real number would probably be +9.5. it's +11.7 overall, but that includes an obviously affected +18.7 from the kawhi injury against the spurs, a result that obviously wasn't going to happen pre-injury.

based on that 1997 jordan thread, i'm starting to think i've slightly overrated him defensively but underrated him offensively. scottie pippen doesn't seem like a #2 offensive weapon good enough to explain the offensive results of the bulls.

Offensive rebounding was pretty much the 96-98 bulls 2nd or 3rd offensive star. Mainly led by rodman

1996 (sansterre data)

Shooting Advantage: +3.3%, Possession Advantage: +5.8 shooting possessions per game (reg season)

Shooting Advantage: +0.0%, Possession Advantage: +9.8 shooting possessions per game (playoffs)

1997 (sansterre data)

Shooting Advantage: +3.7%, Possession Advantage: +3.7 shooting possessions per game (reg season)

Shooting Advantage: +0.0%, Possession Advantage: +7.3 shooting possessions per game (playoffs)

[magicman1978]

c
No. The "Anti-MJ" data consists of 82 games in 1994, 82 games in 1984, 62 games in 1986, and an additional half-season in 1995. His disadvantage stays whether you use regressed data, non-regressed data, srs, or record. As I just illustrated, using regressed data from 1986, 23 year old MJ ends up looking less impactful than Lebron at the ages of 19 and 20. The samples here I'm working off are much bigger than what you're using, per-season and overall. To claim this is all based on a handful of games per season is just wrong.

If your sample for MJ includes 1994 Bulls, do you also include the 2011 Cavs in your same for LeBron?

For your 1986 sample, do you account for his minutes restriction? For 1995, are any considerationsade for him not being in top shape?

[colts18]

If we are going the small sample size, MJ missed 7 games from 1989-1993. The Bulls went 1-6 in those games. Their only win was vs. a Mavericks team that had a 4-50 record. That's not a typo. That was their record. They went on to have a -15 SRS, the worst in NBA history. That's only team the Bulls beat when MJ was out.

[MyUniBroDavis]

If we are going the small sample size, MJ missed 7 games from 1989-1993. The Bulls went 1-6 in those games. Their only win was vs. a Mavericks team that had a 4-50 record. That's not a typo. That was their record. They went on to have a -15 SRS, the worst in NBA history. That's only team the Bulls beat when MJ was out.

I mean I feel the year after they were A 50 win team right? Most of those games came in 1993 and I don’t think the roster changes were that drastic

I don’t think that it’s necessary to the MJ argument that the bulls were a horrible team when he wasn’t there, nor do I think that people argue the lift he provided was the best ever. That team wasn’t built in a way that they’d collapse without him anyways.

It would probably be something along the lines of arguing he has the GOAT offense + value of a elite wing defender situationally and how one views his defense overall.

Again outside of like second stint cavs bron no one in that GOAT offense tier is great on defense and almost all of them have pretty clear issues I think, so it would take him a long way. I have MJs peak at #2 in a tier of its own probably, but I don’t think that neccessarily means that MJ could be put on any team and outdo 3-4-5 in every era or situation even if I think there’s a clear seperation there, I just think he’s clearly better in most


Of course I do think it’s fair to say that if we’re doing something from a perspective of total cumulative impact career value MJ probably isn’t first, but he also retired early after winning the last 6 full years he played in a row, I don’t think it’s a big deal to not hold his longevity against him.

In an case 91-93 is probably closer to bron than anyone else for me in terms of three year peaks (and y’all know how high I rate second stint cavs bron lol). I don’t think any other one comes particularly close at least the way I view it lol. So if you’re low on that stint of bron since alotnof people prefer miami bron I think it’s fair

[Heej]

Finally we have WOWY, which offers us the largest and most inclusive samples of data per-game and per-season and which isn't subject to in-era bias as long as you keep in-era srs fluctuation in mind(The Celtics were the greatest team ever at 50-60 wins a season while the Bulls, Lakers, and Warriors were not close at 60-70 wins a season). And here, when pre-97 greats finally get their shot, Jordan looks significantly worse.

This post so absurd. Do you realize what stat you are referring to? WOWY. With or Without you. Those last two words are key. How can you say MJ look worse in this stat when there was literally no without you's? Michael Jordan had 9 seasons in his career with 82 games. Last I checked, when you play 82 games, there are zero games without you so the stat does not apply at all. What about the 8 years Jordan made an ECF (won 6 titles)? He missed a total of 6 games in those years. How can you talk about a stat when there is literally no sample at all during Jordan's prime? We have just one season at all where Jordan missed significant games. His 2nd season in 1986. After that he missed a total of 6 games in 10 full seasons with the Bulls.

Then you cite PIPM, a stat that never existed during MJ's time.

That dude has polluted all of the Jordan threads with nonsensical intellectualization of biases.

I'm I in the real world? Up until a year ago, MJ was the consensus GOAT who had the consensus peak...cultural impact aside, blah blah blah.

MJ stands alone, always has. It seems like over the last year, there's been this weird wave of folks trying to not just tear down MJ, but remove him from GOAT Convo all together...

Taking it as far as suggesting that people would "probably think" he's the best SG in the league today"

What is happening? Michael Jordan is and has been the greatest player of all time. Nothing's changed.

The only real way for ANYONE to surpass MJ in the public eye is for the Usain Bolt of basketball to show up. And by that, I mean the 6'10 wing who wins 7 rings, and peaks at a level that is clearly beyond Peak MJ, Bron and Shaq.

Until someone is THAT much better, this discussion shouldn't exist.

People are acting like MJs claim to GOAT is nothing but nostalgia, as if we don't have accounts from people who have seen both, and from PLAYERS who have played against Jordan, Kobe and LeBron

That's because this is the 2020s and people haven't been brainwashed by an entire decade of media spoon-feeding people the idea that Jordan was untouchable . People have the ability to collaborate and research and discuss things on their own way more easily now with the internet and many previously accepted cultural norms are being challenged. This post really smacks of old man yelling at clouds, but maybe try to detach yourself a bit emotionally from all this and accept that it's healthy for people to be debating this kinda stuff.

[MyUniBroDavis]

Oh wtf? I misread the thread I thought it was asking what shows he’s the GOAT lol

Candidate is pretty simple right? Are people denying that?

[OhayoKD]

c
No. The "Anti-MJ" data consists of 82 games in 1994, 82 games in 1984, 62 games in 1986, and an additional half-season in 1995. His disadvantage stays whether you use regressed data, non-regressed data, srs, or record. As I just illustrated, using regressed data from 1986, 23 year old MJ ends up looking less impactful than Lebron at the ages of 19 and 20. The samples here I'm working off are much bigger than what you're using, per-season and overall. To claim this is all based on a handful of games per season is just wrong.

If your sample for MJ includes 1994 Bulls, do you also include the 2011 Cavs in your same for LeBron?

For your 1986 sample, do you account for his minutes restriction? For 1995, are any considerations made for him not being in top shape?

2011(pre and post-trade, 15 win pace for the former, 19 win pace for the latter), 2003(17 wins, -9.59 srs over 82 games), 20 win pace in 2018 without love before ty lue is fired, 19 win pace for full season. without samples over miami and both cavs stints, net-rating for both, and Lebron's record without wade and kyrie/wade(60 win pace), without wallace(second best defender) and with shaq(negative) in 2010, lebron leading a +10 psrs team without kyrie and love, ect ect. (NOTE: There's a misconception that players "don't try" when they're tanking but for the most part, players go hard(playing for contracts) and its lineup selection/roster moves where "tanking" happens.)

Corroborated to an extent with consistently hitting the top of the scale with RAPM/RAPM derivatives(but those will ultimately curve down outliers(>25 win lift usually) for regularization

The sample I used for Jordan excludes games where he was on a minute restriction

with an overall improvement of nearly 4 points per game. In his second season, he missed a significant chunk of time after breaking his foot, then logged fewer than 20 minutes in each of his first six games back. Excluding those sub-20 minute games, the Bulls played 15 contests with Jordan at a 40-win pace (-0.3 SRS) that year.

Given the circumstances at play in 95, I just use 96 as a reference making some adjustment for the Bulls adding Rodman to a 50-win team(by srs anyway). If you use the 30 games(over 2 seasons) sample we have rodman in the second three-peat his "Lift" is similar to Durant on the Warriors which I suppose isn't unthinkable given what Falco highlighted earlier:

^ that 2013 is voted as brons peak does not mean that it is. I think ohayo has it either as 09 or 2016

When we talk about Jordan as a GOAT candidate I’m curious if we can differentiate his playoff impact from his RS impact since he’s a playoff rider

For what it’s worth, in terms of relative to competitions def rtg, there are issues with doing it this way but it’s fun and prolly better than the raw values

Postseason offense vs opponent defense (not weighting longer series more)

1991 bulls were +11.5 on offense
1992 bulls were +7.7 on offense
1993 bulls were +10.05 on offense

The magic Lakers from 85-88

1985 +10.9
1986 +7.6
1987 +10.9
1988 +6.8

I’m curious where they rank compared to other playoff title or finals offenses. It’s lower than the 2016 or 2017 cavs and I think the 2017 Warriors off a quick check. Might be fourth

for 2017 warriors, the real number would probably be +9.5. it's +11.7 overall, but that includes an obviously affected +18.7 from the kawhi injury against the spurs, a result that obviously wasn't going to happen pre-injury.

based on that 1997 jordan thread, i'm starting to think i've slightly overrated him defensively but underrated him offensively. scottie pippen doesn't seem like a #2 offensive weapon good enough to explain the offensive results of the bulls.

Offensive rebounding was pretty much the 96-98 bulls 2nd or 3rd offensive star. Mainly led by rodman

1996 (sansterre data)

Shooting Advantage: +3.3%, Possession Advantage: +5.8 shooting possessions per game (reg season)

Shooting Advantage: +0.0%, Possession Advantage: +9.8 shooting possessions per game (playoffs)

1997 (sansterre data)

Shooting Advantage: +3.7%, Possession Advantage: +3.7 shooting possessions per game (reg season)

Shooting Advantage: +0.0%, Possession Advantage: +7.3 shooting possessions per game (playoffs)

Take Pippen and Rodman's raw signals in the second-three peat at face value, and Pippen is making a 50 win(fringe contender) team all-time great and then Rodman is turning an all-time-great team into a GOAT-level team which tracks with what we see in the earlier part of Jordan's prime. Of course, early MJ/the Bulls aren't benefitting from expansion the same way late MJ/The Bulls were(though 90 to 91 playoff improvement can largely be attributed to strong competition turning weak(comparbale by srs with the opponents the 2016 cavs are facing in the east)).

[DraymondGold]

Not sure I have time to reply to all the previous points, but I thought I’d reply to a few of the major WOWY points since you asked in the other thread. Hopefully that's okay!

I moved the discussion here, per your request

Yeah, I'd say if anything, it's the opposite:

Just to reiterate the discussion from the other thread:
-10 year regularized WOWY puts Jordan over Kareem, Duncan, Russell, LeBron. Flat out.
-10 year regularized GPM puts Jordan over all of them but Russell.

It's great that the other players have better raw samples. You can use that to argue for them. But to say that Jordan lacks any comparable impact metrics to those players is just flat out wrong.

Eh, you might be better off responding in the original thread since I already addressed all of this. But whatever, I guess I can just quote the relevant excerpts.

RE: Sample Size

Yeah, no. You're just wrong here. And honestly, since this has now been clarified 3 times and we're literally just looking at what number is larger, I'm surprised we're doing this again:

No. The "Anti-MJ" data consists of 82 games in 1994, 82 games in 1984, 62 games in 1986, and an additional half-season in 1995. His disadvantage stays whether you use regressed data, non-regressed data, srs, or record. As I just illustrated, using regressed data from 1986, 23 year old MJ ends up looking less impactful than Lebron at the ages of 19 and 20. The samples here I'm working off are much bigger than what you're using, per-season and overall. To claim this is all based on a handful of games per season is just wrong.

Otoh, that 72 game you're working with comes out to 7.2 games a season. You can apply as many teammate adjustments as you want, but there's very little to apply them to. Moreover, most of those 72 missed games come from 1986, which as we've covered, taken in isolation, as opposed to being presented as indicative of Jordan's cast throughout his prime, does not paint Jordan favorably. So really, your 10-year WOWYR estimation for prime MJ is basically just based on a season a game.

RE: "raw"
No. See above. It doesn't really matter if it's raw or regressed, if it's not based on a game a season it doesn't favor MJ. Ironically going by SRS instead of record hurts Jordan here.

Re: sample size,

I’m afraid you’re actually the mistaken one

Let’s use Adjusted Plus Minus as a metaphor, since WOWYR is just adjusted WOWY.

Let’s say you want to calculate the adjusted plus minus of a player’s playoffs. Let’s say their team plays 17 games plus one overtime. That’s ~820 minutes total.

Now the star player is on the court for 666 minutes and is off the court for 154 minutes. There’s a few games where they sit out long stretches, perhaps most of the game because they’re injured, and there’s other stretches where they have their normal rotation and only sit out a handful of minutes at a time.

Question: which minutes would you use to calculate their playoff Adjusted Plus Minus? Would you use their entire playoff run, or would you only use the games where they sat out full halves because they were injured?

Every single basketball analyst or statistician would say the former. Now you might counter with: “this shrinks the off sample! Why should we look at their plus minus in games where they only sat for4 minute stints, when we can look at games when they missed entire halves, 24 minutes in a row!”

But that’s not how Adjusted plus minus works. It doesn’t care about your off sample per game. It cares about the total off sample. So if you want to maximize your sample, minimize noise, and maximize your accuracy in measuring the player, you should take the entire playoff run! You shouldn’t throw away games when they only sat for a few minutes in a row. That would be throwing away good data!

This throwing away good data is exactly what you’re doing for WOWYR. Replace 820 total minutes with 820 total games, replace 666 on-minutes with 666 played games, replace 154 missed minutes with 154 missed games. You’re insisting on only using samples like the 82 missed games in 95 for our “off” sample, when we have 154 missed games to work with!

And all this says nothing about the fact that WOWYR sample is getting larger and getting stabler by incorporating the changes in teammates and opponents too. And this says nothing about the fact that I didn't include Jordan's playoff games played or missed, so this larger on-sample of 666 games and the larger off-sample of 154 games could be even bigger.

In short: yes, a 10 year-sample is larger than a single year sample. Yes, a sample of 154 off games (that include 94 and 95!) is bigger than an individual “82 game in 94” sample or an individual “half-season in 1995” sample. And yes, 666 played games is bigger than 78 games in 93 to compare to 1994 or 18 games in 86 or 17 games in 1995. 154 > 82. 666 > 78. 666 > 18. 666 > 17. And yes, this sample disparity gets even bigger considering we’re adding additional data points for his teammates and opponents, unlike raw WOWY.

Again, the fact that the per-season off sample does not matter because that’s not how the calculation works. The sample for WOWYR is just bigger for both on and off games. And, better yet, it's during Jordan's prime during the time period we're interested in, unlike 1984 or 1986.

Re: “It doesn't really matter if it's raw or regressed, if it's not based on a game a season it doesn't favor MJ.”
I’m sorry, but once again, it’s based off 154 game off sample for Jordan, not a handful of games a season. I’m sorry to repeat myself, but when it’s regressed, when it’s adjusted, Jordan looks better. He just does.

Want proof? Here’s the link to the 10-year WOWYR (i.e. adjusted WOWY) page: https://thinkingbasketball.net/metrics/wowyr/. You'll notice Jordan is higher than LeBron James, Bill Russell, Kareem Abdul Jabbar, Tim Duncan, and Wilt Chamberlain. This is the original source for adjusted WOWY. These players aren't tied, Jordan’s number is just bigger. It just is.

RE: Teammate Adjustment
Yeah, uh...

For the sample I think you're using(1987-1997), Pippen only missed significant time(as a starter) in 1989(30ish games) and 1994(10). IOW. this "10 year adjustment" is mostly based on Pippen's exploits as a second-year player.. The "adjustment" being applied to Jordan's ben teammate isn't "correcting" the data, it's distorting it. Not only have we reduced our sample size by a factor of 10, but we're also treating Pippen's exploits as a second-year player as if they are relevant to what Pippen was doing in 1991. (Side-bar: You seem to be including games where rookie Pippen didn't start from 1988 while excluding games where Grant sat in your "without" here. Is there a reason for this?)

This is actually an issue Ben himself outlines(check those articles you linked), and probably why in his own impact evaluations, he mostly focuses on concentrated stretches, not "10-year samples" featuring about a 10th of the relevant evidence. Moreover. even after we've used a 10-year regression to turn a mountain of data into pebbles, Jordan is still well off the very best:

Also, really? Fisher? Again?

Well one, most of the "anti-mj" stuff we've been using uses SRS, not "record". In fact, Jordan generally looks worse when you use srs to assess teams as opposed to a team's record. More importantly, in the specific cases of Lebron and Jabbar, we're not dealing with a "one-off" involving a small sample of games...

somewhat behind the best stuff we see from Hakeem(25 and 30 game lift in 20 game samples in 88 and 90), consistently behind Kareem throughout the 70's(30 win lift in 75, a 29 win improvement with a player similar to oakley as a rookie, 62 wins without his co-star, and takes the depleted remnants of a 30 win team to 45 wins in 77), and a pretty sizable gap compared to Lebron who has multiple 40 win signals for 09 and 10, 30 win signals in his second cavs stint, and is mostly operating at, at least 20+ win lift throughout his prime leading multiple teams to 60 or near-60 win basketball without co-stars on top-heavy rosters(cavs, heatles).

If this was just us looking at the Celtic's one-year turnaround with rookie, I'd be more sympathetic, but as has been covered, Lebron and Kareem's advantage holds throughout their careers at multiple spots in a wide variety of contexts. The gap peaks with the largest possible samples and maintains even in situations where "scalability" theocraticals predict it shouldn't(2015, 2020, 2012, 2005 and 2006). You are not dealing with a few games of Derick fisher, you are dealing with a mountain of data extrapolated from various methods and saying it's comparable (or worse) to grains of sand. If you're that concerned about teammates, why not just look at roster-changes, schemes and adjust as opposed to trying to correct 1991 Jordan's off with a bunch of games in 88without Rookie Scottie.

This equivalency is terrible. Why do you keep pushing it?

A change in win pace when a player plays games vs when they don't is an un-adjusted stat. It's the same as a change in point differential when a player plays in a possession vs when they don't. It's the equivalent of raw plus minus.

Now as I've always said, raw WOWY is more trustworthy than raw on/off. But! There's still useful information in adjusted WOWY. And there are reasons why we would want to adjust raw plus minus data (I thought this was full-consensus? maybe not...)

And again per the source of Adjusted WOWY above, Jordan turns out better. If you disagree, please click the link above -- it's the original source for adjusted WOWY.

But more broadly, why do I keep pushing this equivalency? Well, I've never had to convince someone why they should trust Adjusted Plus Minus more than raw plus minus before. If you don't, that's fine I guess, but I'd reiterate that basically every single basketball analyst and statistician in the world would disagree, including the vast majority of this board. In short: yeah, adjusting plus minus metrics is important. And if you're not convinced of that by the fact that 2001 playoff Derrick Fisher's raw +/- is better than 2016 playoff LeBron's, I'm not sure what will convince you. I guess it's okay to disagree!

Finally...

Just to reiterate the discussion from the other thread:
-10 year regularized WOWY puts Jordan over Kareem, Duncan, Russell, LeBron. Flat out.
-10 year regularized GPM puts Jordan over all of them but Russell.

It really doesn't:

It really does.

Here's the link to the adjusted WOWY database: https://thinkingbasketball.net/metrics/wowyr/ You'll note that Jordan's over all the players I mentioned.
Here's a publicly-available link to Jordan's all-time adjusted WOWY rank: https://backpicks.com/2018/04/08/backpicks-goat-2-michael-jordan/
Here's a publicly- available link to Russell's all-time adjusted WOWY rank: https://backpicks.com/2018/04/02/backpicks-goat-3-bill-russell/
The adjusted-WOWY rank is listed in the top right of the opening graphic, while the un-adjusted-WOWY rank is in the middle right of the graphic.

You'll note that, in adjusted WOWY, Jordan's all-time rank is 4th while Russell's is 30th. 4th place is indeed better than 30th.

Your comment that margin of victory might underrate Russell given the small league is an interesting one!

You say "interesting", but I'd say its "essential". Unless you're more interested in regular season win totals than championships(or championship probability), saying that WOWYR puts Jordan over Russell(and by extension, Wilt) is just misleading. WOWYR says Russell won 11 rings with 35-win help before adding that Wilt was a relative peer. Jordan isn't there. Use large samples, and Jordan looks well off the big-three(Lebron, Bill, and Kareem) while also getting hounded by Duncan, KG, Hakeem, Walton, Shaq and so on. It's not GOAT lvl data, at least not by the standard we seem to be using with Duncan or Hakeem.
Already got into the weeds of how flexible or loose we should apply the labels of "impact", but you're not really getting any sort of favorable comp to the 3 in question unless you go off a few games spread over several years or box-stuff(aupm, pipm, and RAPM exclude the other two and favor lebron in most frames(well PIPM favors Lebron in pretty much any frame), and then WOWY and its derivates favor the other two when a substantial set of data is used(>10 gms/season is my bar)).

Again though, maybe just take it to the other thread where this was all covered in-depth

Again, per the links above, which connect to the official database for adjusted WOWY stats, Jordan looks better than the players you mentioned (Lebron, Bill, and Kareem).

And as I've said already, 154 off-sample games is bigger than 82 off-sample games, and 666 games played is well clear of any of the on-samples in the single seasons you're citing.

Now I'm all for puffing up Russell. You could argue that Russell's underrated by 30th all time in adjusted WOWY, because of the few teams in his league diminishing his teams' SRS. You could use Russell's raw WOWY to argue him over Jordan too.

All I'm saying is that it would not be unreasonable for an opponent to cite that Jordan's adjusted WOWY numbers are better, which they are. Outright.

The quote you said in the other thread was: "Jordan basically lacks any favorable comparisons to the 3 players you mentioned in "impact" stuff over any comparative frame, let alone a 10 year one.Jordan basically lacks any favorable comparisons to the 3 players you mentioned in "impact" stuff over any comparative frame, let alone a 10 year one." This is outright not true. I've provided links and explanations. Jordan looks better in 10-year adjusted WOWY than LeBron, Russell, and Kareem. He looks better than LeBron and Kareem in 10-year adjusted GPM.

Again, I'm not trying to say the other players don't have an argument or statistics in their favor. All I'm doing is trying to say: there is evidence you might use for Jordan. It's not completely 100% one-sided statistically. There are real impact metrics based on actual impact data that portray prime Jordan as GOAT level over LeBron, Russell, or Kareem.

[OhayoKD]

Again we go...again

Re: sample size,

I’m afraid you’re actually the mistaken one

Let’s use Adjusted Plus Minus as a metaphor, since WOWYR is just adjusted WOWY.

The "metaphor" doesn't work because "lineup-level"(what APM uses) data is completely different than "game-level"(what adjusted-wowy uses) data. "game-level" data only looks at the starting lineup. In other words, if Jordan is starting, there is no "off". If Jordan plays 38 minutes, you don't get "10 minutes" of without, you get "0". Similarly all these "adjustments"(how the Bulls are affected by Jordan's teammates not starting) can only be applied when players are out of the lineup. GPM applies a minute filter, but ultimately you reach the same endpoint. Either the player isn't starting, or there isn't any data. Love ya Dray, but if you aren't even grasping basic terminology from what you're working with(and directly quoting), you should hold off telling me I'm "mistaken".

Yes, a sample of 154 off games (that include 94 and 95!) is bigger than an individual “82 game in 94” sample or an individual “half-season in 1995” sample.

What are you talking about?

No. The "Anti-MJ" data consists of 82 games in 1994, 82 games in 1984, 62 games in 1986, and an additional half-season in 1995. His disadvantage stays whether you use regressed data, non-regressed data, srs, or record. As I just illustrated, using regressed data from 1986, 23 year old MJ ends up looking less impactful than Lebron at the ages of 19 and 20. The samples here I'm working off are much bigger than what you're using, per-season and overall. To claim this is all based on a handful of games per season is just wrong.

291>154. This isn't a matter of philosophy. This is math. But hey, at least we've moved past "you're only using a handful of games!" Progress?

In short: yes, a 10 year-sample is larger than a single year sample

No. 32 games over 2 years is not a larger sample than 32 games over 1 year. Increasing the number of years, without increasing the number of games, doesn't extend your sample, it dilutes it.

I also don't know when you polled "basically every statistician and analyst" on 10-year WOWYR, but it would seem the statistician who created that data set prefers my approach. Those 10-year studies you're defending get one sentence in the write-ups for Jordan and Russell. They get 0 sentences in the write-ups for Kareem and Lebron. And when Ben decides who the "greatest floor raiser ever" is, he doesn't bring up "10-year adjusted WOWY" , he takes a bunch of games from a single season, and then adjusts for context. WOWYR's creator sees value in the data he's using being relevant. You apparently don't.

Yes, a sample of 154 off games (that include 94 and 95!)

Yeah, so...are you sure about that? Ben doesn't explicitly tell us what years he includes and excludes but...

there is indirect evidence for a player when his teammates leave the lineup. Let’s say we wanted to know how much Scottie Pippen contributed to the Bulls +9 point-differential in the early ’90’s. In 1994, when Michael Jordan left the Bulls, we could infer something about Pippen based on the change caused by Jordan’s absence.

So while Bill Russell didn’t miss as much time as Jerry West, there’s a bevy of evidence about Russell left by his teammates and all of the time that they miss over the years.

Ben distinguishes between "indirect evidence" and WOWY and we know Ben isn't including the likes of 2011(Lebron), 1970(Russell). I'm not even sure 1995 qualifies as a "prime" year here. Equally concerining is the fact that the players you're comparing Jordan to like say, Lebron or Russell, have missed almost no time outside of those "indirect samples" which WOWYR isn't including. At best, you're using a much smaller sample, spreading it thin, and then throwing it in a comparison with other players where the sample size is truly 2016 Fisher-esque. At worst, we're going off less than a game a season. Either way, your sample is smaller. That part isn't really up for debate.

And all this says nothing about the fact that WOWYR sample is getting larger and getting stabler by incorporating the changes in teammates and opponents too.

Oh, but "this" was said, you just ignored it(I'm noticing a pattern):

For the sample I think you're using(1987-1997), Pippen only missed significant time(as a starter) in 1989(30ish games) and 1994(10). IOW. this "10 year adjustment" is mostly based on Pippen's exploits as a second-year player.. The "adjustment" being applied to Jordan's ben teammate isn't "correcting" the data, it's distorting it. Not only have we reduced our sample size by a factor of 10, but we're also treating Pippen's exploits as a second-year player as if they are relevant to what Pippen was doing in 1991. (Side-bar: You seem to be including games where rookie Pippen didn't start from 1988 while excluding games where Grant sat in your "without" here. Is there a reason for this?)

This is actually an issue Ben himself outlines(check those articles you linked), and probably why in his own impact evaluations, he mostly focuses on concentrated stretches, not "10-year samples" featuring about a 10th of the relevant evidence. Moreover. even after we've used a 10-year regression to turn a mountain of data into pebbles, Jordan is still well off the very best

Moving on...

Again, the fact that the per-season off sample does not matter because that’s not how the calculation works. The sample for WOWYR is just bigger for both on and off games. And, better yet, it's during Jordan's prime during the time period we're interested in, unlike 1984 or 1986.

Honestly at this point, I shouldn't be suprised, but the focus wasn't "84 or 86" Jordan, it was 1988 Jordan with me very kindly assuming the Bulls didn't get any better after Jordan got drafted(cough Oakley cough). Put another way, I juiced Jordan's data, and he still fell short. And, yes, large 82 indirect samples are useful, at least according to the father of 10-year WOWYR:

If Jordan left and the team remained a +9 team, then it would be fairly safe to infer that Jordan was not the reason the Bulls were +9…which tells us that key remaining players on the team, like Pippen and Horace Grant, were the ones responsible for the large point differential.

Heck, it seems Ben preferred an "indirect" 20 game sample over that "10-year one" when it came to evaluating Lebron. If only he'd remembered to adjust!

Well, I've never had to convince someone why they should trust Adjusted Plus Minus more than raw plus minus before.

But 10-year WOWYR is not adjusted plus-minus. And as we've covered, actual RAPM doesn't favor Jordan(8th in a narrow field, best year is well off several from Lebron). Your analogy falls to pieces with a cursory understanding of what's being used.

Finally...

Just to reiterate the discussion from the other thread:
-10 year regularized WOWY puts Jordan over Kareem, Duncan, Russell, LeBron. Flat out.
-10 year regularized GPM puts Jordan over all of them but Russell.

It really doesn't:

It really does.

Unless you care about rs win totals than championships, it does not. Russell looks much better. Wilt scales. Jordan still loses. "Russell won 11 rings with a 35-win cast" is not a win for the guy who won half as much. It just isn't.

There are real impact metrics based on actual impact data that portray prime Jordan as GOAT level over LeBron, Russell, or Kareem.

There are also "real impact metrics based on actual impact data" that potray Fisher as better than Lebron, 2012 KG as "GOAT-LVL", and Manu as better than Duncan. They key word here was "basically", because if one is determined enough, you can always take the Fisher approach and find tiny islands here and that rebel against the sea. I've specifically thrown in caveats like "generally speaking" and "most comparative frames", and "basically" for this reason. Though, in Jordan's case, even those islands only get you so far, as he still comes up short against Magic, D-Rob, and when you adjust for lower championship tresholds(and yes i'd say not making it is unreasonable if you value titles over regular season wins), he comes up well short of Russell and Wilt.

Like at this point you're neglecting basic addition to defend the samples you're using(I don't even know how you got "handful of games" in the first place actually), you're probably overshooting how much there is significantly(Ben consistently does not include "indirect" evidence in 10-year WOWYR), and even if you weren't, that would not change there's even less to go off for the players you're making a comparison with. Never mind, you trying to appeal to a non-existent consensus WOWYR's own creator probably doesn't support because you apparently don't understand the difference between lineup-level and game-level data?

It just isn't there. If you need to reach this hard just to place Jordan over "some" of the guys he's otherwise below, and you still can't get him to the top, it's probably not worth it.

[ceoofkobefans]

A. Intro and ‘What is GOAT tier’?

Seems people here have been able to stay mostly respectful, so I'll take that sign of progress.

I appreciate this sentiment but if this was your objective, I'm surprised then that you almost exclusively focused on the Lebron stuff. I think this part in particular is relevant when we are talking about of all time tiers:

Agreed, but it's a bit of a bind -- I can ignore all your (and others') comments about LeBron to push the discussion that way, but that would also be poor discussing. The other issue is LeBron's the only standard Mount Rushmore candidate (Russell, Kareem, MJ, LeBron) who's fully in the impact stat era, so if I'm trying to compare Jordan to to other Mount Rushmore candidates to show he's in a similar tier using statistics, I'm pushed towards using LeBron for comparison.

I guess that's all to say, I'm trying not to focus too much on whether LeBron is actually better for prime/peak, just whether Jordan is close enough in prime/peak to be in the same tier (since LeBron is presumably in Tier 1 all time). And if there's other places to shift the discussion towards Russell or Kareem or whoever, I'm open to go there instead.

Keep in mind we don't have the data for players like Kareem, Hakeem, Bill, Magic, Bird, Walton, Wilt, or Russell. Jordan is competing in a very, very narrow field here and still doesn't look the best.

Saying Jordan ranks 3rd or 2nd behind other top 10 candidates in metrics that exclude most of Jordan's contemporary and historical competition doesn't really work as proof he's "GOAT tier" unless we are using Colt's much, much broader standard. One you don't seem to follow considering that you have advocated for the exclusion of players for Duncan and Hakeem. This applies, I think to your assertion of AUPM as definitive proof(more on that later).

You’ve suggested Top 3 in a post-1997 stat isn’t high enough to be considered GOAT-level.

Here’s the problem: there is no absolute-consensus #1 player across all impact metrics, not since 2014, not since 1997, not since 1977, not since 1955, not since 1946, not in peak, prime, or career. There just isn’t.

So if being Top 3 in a post-1997 stat is a disqualifier from being GOAT level in impact statistics, that leaves us with two options:
1. Nobody is GOAT tier in impact stats (since nobody, not one, is Top 2 or Top 1 across *every single* impact metric)
2. We need to lower the threshold for what’s considered GOAT level in impact metrics below Top 2 or Top 1 only

If you favor #1, that’s perfectly fine! The boundaries between what we consider “all time” and what we consider “GOAT tier” can be left up to personal preference. And if you prefer to set it higher than any player has gotten so far, that can be perfectly consistent with your criteria! May Victor Wembanyama finally be the one

But me personally, I prefer to have a handful of players in my GOAT tier. So I prefer #2. To me, it’s okay to be “only” Top 3 all time in a stat we have in the time ranges those stats have and still be GOAT level, so long as you’re consistently near the top (hopefully with lots of Top 5s, Top 3s, and Top 1s) across the array of metrics we have. MJ fits this description, so I consider him GOAT tier statistically.

B. Do box-inputs to plus-minus stats and Box-based stats overrate Jordan?

Additionally, when you are dismissing things that are directly drawn from winning like on/off(that directness is very much the point of impact analysis) as useless while championing crude approximations because they make "corrections", it's probably relevant to consider when these corrections are actually making the data more inaccurate:

In short sample sizes, the box-input corrections to actual on-off-based stats are measurably making things more accurate. The idea that raw plus-minus incredibly noisy is complete consensus throughout the community. Some stats aren't even fully stable in full-regular-season sample sizes. So what do we do with this noise? Well, we could try to apply context (which we should anyway!), but the issue here is we don't know for sure whether the noise is pushing the player's value up or down, or how much the noise is changing the player's value. So these box corrections (like AuPM) can measurably make us more correct in small samples (that is to say, closer to the large-sample actual plus-minus-based value that corresponds to contributions to winning).

As Lebron and Jordan are virtually tied on the offensive portion of all these metrics, simply replacing the defensive component with actual impact data, knocks Jordan off his perch. And remember, this is not including Kareem [streamable][/streamable]whose defenses were 4 points better, or Russell who won the most, by a landslide, on the strength of his team's defense.

BBR BPM is on par with RAPTOR in terms of predictive accuracy IIRC(notably behind direct rapm extraps. like EPM when tech is equalized). As far as BPM is concerned, Jordan is a significantly better defender than Kareem. Simply hedging between defensive impact signals and defensive box-score data knocks knocks Jordan out of range.

That a metric makes adjustments of some kind does not make it inherently better, and proper analysis involves weighing the merits and cons of different metrics and then deciding what adjustments/caveats/context needs to be applied. There is a trade-off here.

You get less noise, but you also get inaccuracy that skews towards a certain archetype. And when the "corrected" data is consistently disagreeing with "real" data, then adjustments should be made. That is the value of "raw" impact. And any credible impact analysis will factor in those types of signals.

The on/off Ben calced is, to my knowledge, the only available sample of data which doesn't utilize an artificial scale and accounts for defensive impact. Don't you think the idea that we shouldn't even consider this while we use metrics that equate steals per game with defense a little silly?

Good points here!

Absolutely, various metrics can tend to favor certain archetypes. There's the famous issue of earlier Basketball-reference all-in-one stats (either PER, ws/48, Basketball Reference's BPM 1.0, I forget which) consistently overrating bigs with assists and guards with rebounds.

Here, I definitely agree that box-only-based stats tend (but not always) to overvalue perimeter defense, particularly steals. I definitely don't have peak defensive Jordan over peak defensive Kareem in absolute defensive value (yay, a non-LeBron goat candidate to talk about! ). A few qualifiers that may shrink the defensive gap between Jordan and Kareem:
(1) One could make some sort of team-building relative-to-position argument in favor of Jordan... i.e. that Jordan's a bigger defensive outlier among guards than Kareem is among bigs, and thus teams get a bigger boost choosing Jordan (because it's easier to replace big man defense). Not sure this is very convincing, but at a minimum I do enjoy these sorts of philosophical discussions about how much to give 'bonus points' for scarcity of value at a position.
(2) I'm also open philosophically to the idea that Jordan's all-time motor leads him to 'get the most' out of his defensive value more consistently as he gets older. Kareem and LeBron both have seasons when they get older where their defense slides behind what they're capable of, maybe to preserve energy for the offensive end, or to preserve energy in the regular season for the playoffs. For example, I'd absolutely take 96/97 Jordan defensively over 2018 LeBron defensively. Kareem also has a reputation for lowering his defensive effort as he got older, though I'm less familiar with how his effort waxed and waned in specific seasons.

To shift our the conversation back to statistics (this thread's focus), a few more qualifiers about why we shouldn't throw out out the box-based statistics:
(1) Even if steals are overrated in box-based defensive measurements (e.g. if the box stats miss the fact that the steals come expense of unnecessary gambling, as they may for Jordan), steals are still individually the most valuable defensive play someone can make. A guaranteed change of possession and high probability of starting a fast break (the best type of team offense) is super valuable.
(2) Some of the better box-stats I'm using have input weights that are calibrated overall, rather than calibrated for offense and defense separately. In other words, we calculate those stats as a whole, then split into defensive and offensive components after. This may effect the error: in this case, a player that’s overrated on defense may actually be underrated on offense (since the mistake is whether we attribute the value to offense or defense, not in the overall value). Not a guarantee, just a possibility. Like you say, it may also mean they’re overrated overall.
(3) The fact that stats overrate certain archetypes is why I tend to like looking at a variety of stats across the board (e.g. including whatever information we have from stats that are almost all value-based like RAPM/PIPM, hybrid stats like AuPM, WOWY-based stats, etc.). So if a player looks pretty good across the board, e.g. if he almost always places in the Top 5 ever, that's a pretty compelling signal to me.

C. WOWY vs WOWYR: Jordan, LeBron, Russell, Kareem

I also think there are some basic inaccuracies here that we should address, acknowledging I was under the incorrect impression pre-97 aupm also lacked plus-minus data.

Happy to talk more! I certainly could be incorrect in my understanding in other areas

That being said...

...uh, no

I’ll see your “uh, no” and raise you one “uh, yes”!

WOWY is short for “with or without you”.
WOWYR is short for “With or without you, regressed”

The process that changes raw plus minus data to Adjusted Plus Minus (APM) is a statistical procedure called a regression.

You’ll note that regression (which again is the “Adjusted” in Adjusted Plus Minus) is right in the name of ‘WIth or Without You Regressed’. It’s right in the name!

So again: WOWY ~ Raw Plus Minus. WOWYR ~ Adjusted Plus Minus.

You can read more about how WOWYR is calculated here: https://thinkingbasketball.net/metrics/wowyr/. You’ll note this is the original article that published WOWYR, and it mentions the parallel of WOWYR and APM in the first paragraph.
You can read more about how APM is calculated here: https://www.nbastuffer.com/analytics101/adjusted-plus-minus/
[Note: Since this is a confusion I often make myself, a regressed/ “Adjusted” data is not the same as regularized from RAPM. The first just corrects for the context of other players, to isolate for the impact of individual players. Regularized is when we try to correct for outliers / noise to make the measurement more consistent. The first is a must-do if you want to be sure you’re looking at the impact of an individual player specifically. The second may be beneficial, but there are different regularizing schemes and some of the outliers they correct for may occasionally be actual signal, not noise]

Instead of using results from lineups within a game (play-by-play data) like traditional APM, game-level plus-minus uses final scores from game to game for the players from that game. This allows for a historical, apple-to-apples comparison of per-game impact from before play-by-play was available (1997).

As I covered before, the "correction" is marginal. But frankly WOWY and WOWYR doesn't really make a difference here, because as long as you are using large samples, even corrected impact still has Lebron consistently looking better:

Before Michael, the 1984 Bulls were a 27-win team (-4.7 SRS) with an average defense and a futile offense that finished 5 points worse than the league (rORtg). Jordan immediately provided the scoring punch that they needed and Chicago improved to just above average on offense in his rookie year, with an overall improvement of nearly 4 points per game. In his second season, he missed a significant chunk of time after breaking his foot, then logged fewer than 20 minutes in each of his first six games back. Excluding those sub-20 minute games, the Bulls played 15 contests with Jordan at a 40-win pace (-0.3 SRS) that year.

The ’06 Cavs were even more impressive, thanks to a breakout year from James. With Ilgauskus and Gooden now accompanied by Larry Hughes (a moderate creator and inefficient scorer), the offensively-challenged Snow and two shooters (Donyell Marshall and Damon Jones), Cleveland churned out a 5.1 SRS when healthy (56-win pace) with a +6.6 offensive efficiency in 30 healthy games. A similar rotation ticked along at a 51-win pace in ’07 (3.4 SRS) in a larger sample, but the offense regressed to near-average, meaning the ’06 result was likely an aberration. (LeBron’s offense regressed slightly in ’07 too, likely contributing to the backslide.) Still, the period demonstrated that pre-prime LeBron-ball could buoy offenses while stuffing the court with defenders and a few shooters.
(The cavs were a -9 srs team before drafting Lebron)

Jordan also lags behind Kareem in larger(>10 games per season) samples, and Russell. and Hakeem. The most relevant part of WOWY vs WOWYR here is the inclusion of 82 game stretches in impact analysis. And I think you and I can both agree that full season samples can be very, very useful, even if there's noise to account for. The only thing that has Jordan comparing favorably(to lebron, not everyone in history) is 10-year data, but again, let's consider the sample in question:

Your 10 year-set is taking 2.2 games per season from Russell and then throwing less than a game a season for Jordan alongside a larger sample from one season that does not compare favorably with Lebron, Kareem, Hakeem, Bill, or various other players I've neglected to mention(KG, Shaq, ect). And for all that, if we account for certain eras requiring lower SRS for high championship probability...

At the height of their dynasty, the Celtics were comically dominant. From 1962-65, their average margin-of-victory (MOV) was over 8 points per game. During the same time span, only two other teams even eclipsed 4 points per game – the ’64 Royals and the ’64 Warriors. And all of Boston’s separation was created by its historic defense, anchored by Russell:

...Jordan is still well behind Russell(and by extension Wilt):

Notably, if we take WOWYR seriously, Bill Russell led the greatest team ever with 35 win help throughout his prime while Jordan barely won half as much with 40-50 win help. While Jordan looks marginally better than Lebron, he's not really within range of GOAThood.

Just to give Jordan the slightest empirical advantage over Lebron, we reduced our per-season sample by a factor of ten and Jordan still comes out well behind Russell and Wilt(as well as Magic and D-Rob).

Regressed or non-regressed, if we use larger samples, Jordan plummets relative to Lebron and Kareem. We can discuss the merits of this type of data, but trying to paint his WOWYR stuff as some GOAT-lvl indicator doesn't really work. Notably, the disparity generally comes from the defensive side of things. A disparity we shouldn't be surprised box-stuff can't account for. It's also a disparity many of the theoretical excuses made for the lack of comparable influence doesn't really consider(The 2015 Cavaliers say hello).

Lots of interesting thoughts here! Let me try to organize them into a few main points, but do let me know if I missed any!

Claim 1: “Regressed or non-regressed, if we use larger samples, Jordan plummets relative to Lebron and Kareem. [and Russell, etc.]”.
So right off the bat, this is not true. WOWYR looks at 10-year samples, and Jordan comes out ahead of LeBron, Kareem, Russell, and all the other players you mentioned. In GPM, again using a 10-year sample, only Russell comes out ahead, while Jordan comes out ahead of LeBron, Kareem, Wilt, Hakeem, etc.

You provide lots of interesting raw data! And for the record, I do find the raw data quite compelling for people, Russell in particular since our stats for him are so limited. There’s a reason I have Russell in my GOAT tier!

But remember, these stats are “UnAdjusted”…. and unadjusted plus minus data can make 2001 Playoff Derrick Fisher look better than 2016 Playoff LeBron. They make no corrections for opponent (they use Margin of Victory, not SRS), nor teammate health, nor opposing health, nor overall team context. So if the Adjusted data disagrees with the raw data, I’m personally more inclined to believe the Adjusted data, or at least heavily consider its validity.

Claim 2: “WOWYR is based on small individual sample sizes, unlike single samples of raw WOWY in longer periods where players miss time”.
For example, you cite that WOWYR only uses 2.2 games per season to count Russell’s “off” sample, and imply there’s a similar minuscule number of games to calculate Jordan’s off sample. I’m not sure this is true.

Remember, WOWYR is Adjusted for teammate context, and so it uses the WOWY of every other major player on the team to help isolate Jordan’s. The graphic in the WOWYR article I linked above shows this.

In other words: Jordan’s numbers aren’t just based on the 666 games he played and the 154 games he missed in the 10-seasons we’re looking at (154 -> 72 games out if we ignore 94). We can also use the games when other players were in or out of the lineup: Pippen was out for 113 games in this timespan, Rodman was out 18 games, Horace Grant was out 28 games, Bill Cartwright was out 66 games, Luc Longley was out 102 games, John Paxson was out 86 games, early guard Sam Vincent was out 81 games, early forward Gene Banks was out 19 games, early center Earl Cureton was out 39 games [at least by my count]

… every one of these games with lineup changes goes into the Adjusted calculation for WOWYR. And every one of them is used to tell us a little bit more about Jordan’s specific value. Now it’s not as much as we would need to decrease the WOWYR noise to as low as we’d get for RAPM, but that seems like a lot of games to me!

Regardless, it’s certainly a heck of a lot more than just using a handful of games a season to find Jordan’s raw WOWY. This kind of increased sample size would also apply when calculating WOWYR for players like Russell, Kareem, LeBron, etc. That’s one of the benefits of WOWYR over raw WOWY… the fact that we’re adjusting for teammate-context allows us to use the teammates’ games in and out of the lineups to increase our sample size pretty dramatically.

(2) The effect of good coach on WOWY: WOWY is sensitive to coaching. If a player is missing one game, for example, a poor coach may put less effort into adjusting the game plan than if a player is missing a lot of time in a row. If one coaching staff is much better than another coaching staff, the better one might do a much better job at filling in when a player is missing than a bad one. Jordan had great coaching with Phil Jackson. This is the kind of thing that would limit his raw WOWY.

The assumption that good coaching must depress impact is questionable. In fact I would say bad-coaches failing to optimize a star player's influence is as common than the opposite. I'm also not sure why you're using a one-game sample in your analogy when its multiple 82 game samples that are being relied upon the most here. If your theory holds, then the relatively pedestrian stuff we have under Jackson's significantly worse predecessors should be sparkling. Moreover, if you're concerned about coaching ajustments, then WOWY is really the way to go here, as it's much easier to make adjustments with some pre-time before a game or a season, than it is to make adjustments when a player leaves half-way through.

Frankly this is an exceptionally weak approach to take with say, Lebron, considering that Lebron looks better in larger samples of "off", and his teams tend to look the worst when the team is given time to adjust. This includes his time under Erik Spoelstra where the heatles without Lebron did not look as good as the Bulls without Jordan or the Bulls without Jordan and Grant(at least by SRS).

Not to be too critical here, but this seems like another example where you've come up with a seemingly viable theory, without actually looking where the breadcrumbs lead.

WOWY(and WOWYR) is indeed noisy, which is why it's good to look for replication across a variety of contexts:

somewhat behind the best stuff we see from Hakeem(25 and 30 game lift in 20 game samples in 88 and 90), consistently behind Kareem throughout the 70's(30 win lift in 75, a 29 win improvement with a player similar to oakley as a rookie, 62 wins without his co-star, and takes the depleted remnants of a 30 win team to 45 wins in 77), and a pretty sizable gap compared to Lebron who has multiple 40 win signals for 09 and 10, 30 win signals in his second cavs stint, and is mostly operating at, at least 20+ win lift throughout his prime leading multiple teams to 60 or near-60 win basketball without co-stars on top-heavy rosters(cavs, heatles).

The disparity is consistent. I don't need to cherrypick one year or approach to observe a gap. That's a pretty good indication that this can't just be put down to "noise". It's good to look at everything and assess the evidence holistically. The best possible signal I can get for Jordan is to take record instead of srs for 1986, ignore 84, **** with the minute thresholds ,ignore Ben's much more pedestrian appraisal, pretend Oakley didn't help them defensively, and you get 32 win-lift for a sub 50 win team. That took many, many extra steps and it still doesn't get you to what Lebron does in 2009, 2010, or 2015 and 2016. That 23 win-appraisal I throw around works on the assumption there was no improvement after MJ was drafted(again, ignoring evidence that Oakley helped alot defensively). "Impact" is just not a winning case unless you ignore the forest for trees.

Re: coaching, I’m not married to the theory. But let’s return to 2001 playoff Derrick Fisher vs 2016 Playoff LeBron. Raw plus minus says Derrick Fisher is better than LeBron. However, when we adjust for teammates and context, we come to the (obvious conclusion that) playoff LeBron is actually better. If we’re tight on time, we can trust that the math works out that way and move on, or if we’d like to dive a little deeper, we can start to contextualize what made Derrick Fisher’s raw numbers mislead us. The context I’d personally apply is the fact that his minutes aligned with prime Shaq, Kobe, and a variety of other talented teammates.

Now we return to WOWY and Jordan. Raw wowy suggests Jordan is worse than some of his all-time competitors. But when we adjust for the context (exactly like we did above), we find that Jordan is suddenly more valuable than LeBron and Kareem and many other all-time players in their 10 year span. The coaching was just my attempt to explain why Jordan’s raw data might underrate him, just like I attempted to propose context for why Derrick Fisher’s raw data overrated him. You might believe the argument, you might not, you might have other ideas (I’m all ears!)… but these contextual arguments don’t change the fact that, mathematically, when we Adjust Jordan’s raw WOWY data, he ends up on top over the other GOAT candidates.

Now you might make arguments for why those other players are still better than Jordan. You might cite the nosiness of WOWY-based data. Your comment that margin of victory might underrate Russell given the small league is an interesting one! But to return to the premise of the thread: WOWYR is another impact stat that portrays Jordan as GOAT level, over Russell, Lebron, and Kareem.

D. RAPM: Jordan, LeBron

Speaking of which...

We have less than 20% of Jordan's games measured in Jordan's 6 best years, we have strong evidence that the games we do have underrate Jordan, and Jordan nonetheless comes out 8th all time, tied with peak 2013 LeBron. Again, LeBron's non-peak years may still end up being higher than Jordan.

As far as the data you're actually using is, 2013 Lebron is not Lebron's peak, and the comparison here is Jordan vs Lebron, not "Jordan vs Miami Lebron" or "Jordan vs whatever year of Lebron might give MJ a semblance of a case". You are more than smart enough to recognize the difference between letting the evidence speak and strangling it so that it fits your priors.

You also neglect to mention that the data we have for 1988 actually skews in Jordan's favor as the Bulls did worse during the portion of the season we don't have data for. As I'm sure you're aware, there's plenty pointing to 1988 as Jordan's most "imapctful" season and that conclusion would actually fit the "bad coaching" theory you offered earlier. It is fair to point out uncertainty, but trying to take data that clearly favors Lebron as actually "pro-mj" because more data may improve how Jordan looks is a bit of a leap. As it is, we do have playoff on/off here, and it doesn't support that conclusion. Jordan's on/off arcs downward(in line with a defensive decline observed in both Blocked and Ben's film-tracking) from 1988 to 1993 before rebounding for the second-three peat. As it is, Jordan's 1988 also scores near the top in the offense-skewed stuff you seem to prefer, so this honestly seems like a questionable prediction.

Also important to note, before we use "tiers" to explain this away, this data only really exists for the peaks of post 1997 greats(and MJ), so Jordan only looking sizably worse than one modern player(doesn't really look better to me than duncan or kg here though maybe an expert like Jaivi can offer some distinction), doesn't mean he's "close" to being "the greatest". He flatly doesn't score close to Lebron here, and we have no way to know if that would apply to players like Wilt, Russell, or Kareem. Notably, RAPM consistently places Lebron well ahead of the likes of KG, Shaq, and Duncan(I recall seeing a 5 year average where the gap between lebron and 2nd place KG was similar to the gap between KG and 7th place Nash), players who, with more "apple to apple" pure impact analysis look quite comparable to Jordan. Crude comparison, but at this point it's a straw on a camel that broke yesterday. It's not as emphatic as WOWY(regularization will do that), but it's just not a winning case for puffy-j.

Good points here. A few thoughts.

Re: “2013 LeBron is not LeBron’s peak”, most people consider LeBron to have peaked in 2013. You can see in any past Greatest Peaks project on this board and find 2013 LeBron voted in over other years. Look on other sites and other talk shows and people will normally say 2013 LeBron. This brings us to the question of why RAPM rates 2013 Lebron lower than other years, or whether it’s missing anything. Perhaps the larger group thinks there’s a greater playoff resilience in 2013 than in 2009? Or perhaps the larger group is just mistaken? One possible explanation for why 2013 is lower is that RAPM measures impact in a specific role… perhaps LeBron’s role in 2013 was less suited to maximize his individual value. (You’ll note that this is the exact argument made against LeBron by scalability proponents).

Re: 91 Bulls vs 88 Bulls: The 1988 Bulls over performed by 1 single game: they were on a 51-win pace in our sample and they won 50 games in the full season. The 1991 Bulls underperformed by at least 8-9 games: they were on a 52/53-win pace in our sample and they won 61 games in the full season. They were on a 64-win pace by SRS. 1988 is well within random variance; 1991 Bulls is the largest underrated sample in the entire Squared2020 database.

You can argue that RAPM still favors LeBron. Personally, I’m open to the idea. But since the thread asked “are there any impact metrics that can argue Jordan is GOAT-tier”, it seems reasonable to say that a Top-8-all-time number in a sample that drastically underrates how good the 91 Bulls were and doesn’t measure Jordan’s value in his other peak years of 89 or 90, or other near-peak years of 92 or 93 might suggest GOAT-tier or near-GOAT-tier impact.

Re: definition of GOAT tiers, see my first comments above In short, perhaps we just have different definitions of what GOAT-tier is. Which is okay!

E. Is Duncan GOAT-tier??

To me, a more fair characterization of the box-approximations of PIPM and RAPTOR is that they predict true plus-minus-based PIPM and RAPTOR, with wider error bars than the plus-minus based ones.

Sure, but it's not just "wider error bars", it's "wider error bars largely because they ascribe outsized(relative to historical precedent and actual impact signals) defensive value to smaller steal and block accumulators." But even then, looking at the metric that accounts for defense best...

And that if Jordan looks comparable to LeBron

But he doesn't. You specifically chose a favorable frame of comparison for Jordan(3-years consecutive), and Lebron has, not one, but two better stretches when we utilize that frame. Going off the data RK listed, Lebron has the 2 highest scoring years(with 2009 being far ahead of anything else), and 5 of the best scoring 7. I could literally chuck the best scoring year by far, and Lebron would still look better. Jordan does not look comparable, and he does not rank 3rd-all time, he ranks 3rd among the players we actually have data for. PIPM dates back to 1977. That leaves at least 2 players with consistently better impact indicators completely out of the room.

Agree to disagree here. I’ve provided plenty of evidence for why I they’re comparable statistically, and you’ve provided plenty of reasoning against it. It’s okay to not always agree with everyone!

As for AUPM, you can shake off Lebron if you use a three-year frame(note I said "generally speaking" and "most comparative frames" as qualifiers), he still falls short here to Duncan. Considering that AUPM is constructed as a combination of on/off and BPM, that Duncan grades out #1 here is rather impressive, especially since we are using a frame of comparison(3 years consecutive) that gives him the best looking case. And remember, Jordan does not score "2nd all-time", he scores "2nd among a minority of historical players in this specific metric using the most favorable possible comparative lens". Considering you don't have Duncan to have "GOAT tier" impact indicators, it seems logically inconsistent to me to argue that impact-data potrays Jordan as a "definite GOAT-tier player"

...Duncan scores higher in aupm despite aupm being partially constructed with BPM, scores as high as a pretty optimistic MJ WOWY appraisal in injury plagued 04/05, looks similarly dominant in RAPM stuff(though this gets very noisy with different scales), and won 57 and 62 wins at his most valuable looking stretch as opposed to 50 for Jordan in 1988.

Hakeem looks better if you use his very best WOWY samples, looks better in his first three years, and looks similarly impactful throughout his prime, while scoring higher in postseason PIPM(the box metric which most closely is tied to actual defensive impact.(remember that pre 97, none of the "plus-minus" stats you reference have plus minus(or film tracking)). Hakeem also scores similarly in 97/98 on/off despite arguably being further from his peak than Jordan was those years.

I don't mind different definitions, but I think its a good idea to keep our thresholds consistent. When you tell me someone has GOAT-tier impact stuff, I want to see something that suggests you were the greatest. Maybe you're using a more liberal definition, but I don't think consistent application leaves Jordan significantly above TD and The Dream.

I think this is flat-out untrue. Jordan's playoff Augmented Plus Minus, based on actual plus minus data, is better than LeBron's. His WOWYR is over LeBron/Russell/Kareem. And all the approximations of more accurate stats we have show him as GOAT tier.

In AUPM, Jordan looks worse than Duncan and better than Lebron in one framework while looking worse than LBJ in most others. That is also just a fraction of nba history being accounted for.

His WOWYR is flatly worse than Russell's(and Wilt) over 10 years(when we adjust for lower srs-championship tresholds), and when we take >10 or 82 game samples instead of a sample of 6 games over 8 years, Jordan scores well behind whether you prefer WOWY or "corrected WOWYR. I also don't know what you're basing these metrics being "the most accurate" from. The box-stuff specifically gets less noisy by chucking out defensive accuracy and the more useful method(imo) where we just replace the defensive box-score with defensive impact, immediately sees MJ plummet.(Jordan is tied or ahead of Lebron in D-RAPTOR, ahead of Kareem in BBR D-BPM, well, well behind on D by basically all impact stuff).

Perhaps these stats aren't as bad as PER, but nonetheless, they skews heavily towards Jordan(at least relative to the history of great defenses, and the "real" impact signals of the players in question), and Jordan still does not get #1 if it has actual on/off or plus-minus. Coincidentally, his actual on/off looks much, much worse, as does WOWY and adjusted WOWY over serious samples(>10 games).


If you loosen you definition of Impact(non-plus minus RAPTOR and pure Box with weak correlates) you can get Jordan there(along with someone like Duncan), but I feel my definition of "impact" is more in spirit with what impact denotes and is ultimately more useful.

Accept or reject that, but consistency is key:

...Duncan scores higher in aupm despite aupm being partially constructed with BPM, scores as high as a pretty optimistic MJ WOWY appraisal in injury plagued 04/05, looks similarly dominant in RAPM stuff(though this gets very noisy with different scales), and won 57 and 62 wins at his most valuable looking stretch as opposed to 50 for Jordan in 1988.

Hakeem looks better if you use his very best WOWY samples, looks better in his first three years, and looks similarly impactful throughout his prime, while scoring higher in postseason PIPM(the box metric which most closely is tied to actual defensive impact.(remember that pre 97, none of the "plus-minus" stats you reference have plus minus(or film tracking)). Hakeem also scores similarly in 97/98 on/off despite arguably being further from his peak than Jordan was those years.


IIRC, you have dismissed both Duncan and Hakeem as having GOAT-level data on multiple occasions. If Jordan's impact stats potray him as "absolutely GOAT-Level at his best", why don't you extend that for Hakeem and Duncan who do just as well if not better using data which actually has "impact" in it.

Is Duncan a GOAT candidate according to "impact"? If so, sure, put Jordan there. If not, then I don't think MJ really has a case(at least if "impact" is the lens).

And yes this post was brought to you by the San Antonio Spurs

Love the last comment!

I'm open to suggestions that Duncan's peak is GOAT tier, at least statistically. I personally have Duncan a touch below ~5th GOAT peak, but he does have a good high-end argument. Duncan's also great on the longevity front. My concern for Duncan is the 'width' of his peak and his overall prime-ability being a touch below the other GOAT-tier players.

If I'm trying to measure a player statistically, I like to look at a variety of stats. Duncan's plus-minus based stats are indeed GOAT level.But Duncan grades a bit lower in box-based stats and WOWY-based stats (still great all-time, but under all the GOAT-tier players we've been discussing).

I've pretty thoroughly discussed Hakeem's lower performance in the statistics in other threads, not looking to get too far down that rabbit hole here.

Not responding to the other things atm but while yes WOWYR and GPM Tries to adjust for other teammates missing iirc the bulk of the calculation is based on that players WOWY sample so still the WOWY sample being small still matters (there’s a reason why John Stockton is 2nd all time in WOWYR and it’s not because he has a goat argument). I mean WOWYR data still takes MJ over guys like Russell and Kareem so take that as you will (also anyone else mentioned other than LeBron i believe).

[DraymondGold]

Part 1: WOWY and WOWYR Discussion

Again we go...again

Re: sample size,

I’m afraid you’re actually the mistaken one

Let’s use Adjusted Plus Minus as a metaphor, since WOWYR is just adjusted WOWY.

The "metaphor" doesn't work because "lineup-level"(what APM uses) data is completely different than "game-level"(what adjusted-wowy uses) data. "game-level" data only looks at the starting lineup. In other words, if Jordan is starting, there is no "off". If Jordan plays 38 minutes, you don't get "10 minutes" of without, you get "0". Similarly all these "adjustments"(how the Bulls are affected by Jordan's teammates not starting) can only be applied when players are out of the lineup. GPM applies a minute filter, but ultimately you reach the same endpoint. Either the player isn't starting, or there isn't any data. Love ya Dray, but if you aren't even grasping basic terminology from what you're working with(and directly quoting), you should hold off telling me I'm "mistaken".

Re: "isn't any data" ...

what? What do you mean there isn't any data? Of course there's data. The "lineup level data" is based on who played the game or not. When the player is playing, that's there on sample. When they aren't playing, that's their off sample!

When we look at raw on/off, we look at the possessions when a player's on and compare them to the possessions when the player's off.
When we look at raw With/Without You, we look at games when the team played 'with' the player and compare them to games when the team played 'without' the player.

When we want to find a player's adjusted plus minus, we take the raw on/off of them and all their teammates and opponents and apply a regression.
When we want to find a players' adjusted WOWYR, we take the raw WOWY data for them and all their teammates and opponents and apply a regression.

We don't only look at the "on" of on/off and say there "isn't any data" for the off sample. It's called on/off! You need an "on" part and an "off" part. You need a "plus" part and a "minus" part. Likewise, for WOWY, you need a "with you" part and "without you" part.

You do this for every player, then apply the adjustment. Of course there's data for the on and off sample!

re: you should hold off telling me I'm "mistaken", I only said that in reply to your comment of "You're just wrong here." But I'm happy to leave that style of banter out if you'd prefer -- definitely don't mean to offend

Yes, a sample of 154 off games (that include 94 and 95!) is bigger than an individual “82 game in 94” sample or an individual “half-season in 1995” sample.

What are you talking about?

No. The "Anti-MJ" data consists of 82 games in 1994, 82 games in 1984, 62 games in 1986, and an additional half-season in 1995. His disadvantage stays whether you use regressed data, non-regressed data, srs, or record. As I just illustrated, using regressed data from 1986, 23 year old MJ ends up looking less impactful than Lebron at the ages of 19 and 20. The samples here I'm working off are much bigger than what you're using, per-season and overall. To claim this is all based on a handful of games per season is just wrong.

291>154. This isn't a matter of philosophy. This is math. But hey, at least we've moved past "you're only using a handful of games!" Progress?

I've never been dealing with a handful of games, that was your assertion mon ami I've always been dealing with an off sample of 154 games. But agreed, it's good to be on the same page about what we're both saying And yes you're right if you sum your entire off sample, you do get a bigger off sample

But for the record, 666 + 154 games > the sample you're dealing with. Your on sample consists of 82 games in 85 (to compare to when he was out in 84), 18 games in 86 (to compare to when he was out in 86), 82 games in 88 (to look at Jordan pre-Pippen), 78 games in 93 (to compare to when he was out in 94), and 17 games in 95 (to compare to when he was out in 95).

WOWYR's sample of 666 "on" games > your 277 "on" games. WOWYR's total sample of 820 total wowy games > your sample of 568.
The "on" sample I'm using is 2.4x bigger, while the "off" sample you're using is only 1.9x bigger. And to reiterate, the on sample your using doesn't use any data from 1989-1992, when most people consider Jordan was at his best. Meanwhile, the adjusted WOWYR sample includes both that data for the on sample as well as teammate WOWY data from that time period to better pin down Jordan's value.

In short: yes, a 10 year-sample is larger than a single year sample

No. 32 games over 2 years is not a larger sample than 32 games over 1 year. Increasing the number of years, without increasing the number of games, doesn't extend your sample, it dilutes it.

. First off, that's not how WOWY or plus minus work. To calculate adjusted plus minus, you 1) calculates the "on" part over some time span, 2) calculate the "off" part separately over the same timespan, 3) calculate on/off = on - off, 4) Adjust to get adjusted plus minus.

Each on/off part is calculated independently. So if you add years, that gives you more data! No dilution. Just more data points. We have more data points for the "on" sample. Still a good amount of data points for the "off" sample. And we have a comparatively massive amount of datapoints to adjust for his teammates' contributions, while raw WOWY has exactly 0 datapoints to adjust for teammates.

But second off, again your sample ignores almost all of Jordan's best years. Adding more data points to adjusted WOWYR somehow dilutes it but ignoring 1987, 1989, 1990, 1991, and 1992 for Jordan doesn't dilute it? This just doesn't seem consistent.

I also don't know when you polled "basically every statistician and analyst" on 10-year WOWYR, but it would seem the statistician who created that data set prefers my approach. Those 10-year studies you're defending get one sentence in the write-ups for Jordan and Russell. They get 0 sentences in the write-ups for Kareem and Lebron. And when Ben decides who the "greatest floor raiser ever" is, he doesn't bring up "10-year adjusted WOWY" , he takes a bunch of games from a single season, and then adjusts for context. WOWYR's creator sees value in the data he's using being relevant. You apparently don't.

Re: every analyst, yeah I've never heard of a single analyst preferring raw plus minus to adjusted plus minus. There are surveys out there asking NBA organization members which (publicly available) stats they use, and they all pretty universally the regressed data.

Re the creator of WOWY, when Ben looks at Jordan's WOWY signal... including his individual samples of raw WOWY, his full-prime raw WOWY, and his full prime adjusted WOWYR... Ben concludes that there's a clear statistical argument for Jordan to be GOAT level (where GOAT level is Top 4 ever).

That he cites the individual samples of raw WOWY more often in his articles may just be a function of simplicity and communicability... it's quicker to explain a single sample of raw WOWY than it is for him to explain individual samples of raw WOWY -> full prime raw WOWY -> adjusted WOWYR, especially for an audience unfamiliar with his work.

Re: "you apparently don't.", here was the original Premise of this discussion:
Q - are their actual impact metrics that portray Jordan as GOAT level?
Me - yes, here's links to a whole bunch of them. [with GOAT level being Top 4 ever]
You - No, there are no impact metrics that portray Jordan as GOAT. The ones you cited aren't true, and the box ones don't count
Me - Yes the ones I cited are true, here are yet more links to prove this. Again, I'm just saying there are stats that can be used to argue Jordan's Top 4 ever, not that all stats universally favor Jordan
You - "You apparently don't see" the "value in the data he's using as being relevant"
Me - ???... What ?

EDIT: I wanted to provide quotes for when Ben actually outright says Jordan's WOWY and WOWYR data can be used to argue he has the GOAT peak and prime And he does indeed cite adjusted WOWY: "The adjusted game-level data we have on Jordan echoes the common sentiment that he’s one of the most valuable players ever; he’s right at the top of these three studies with an average per-game value of +8.2"

And just a few lines above, when talking about Jordan's individual raw WOWY sample when returning in 95, he says: "lifting slightly above average offenses by 5 or 6 points is GOAT-worthy". So not only does the creator of the stat use my approach too, but he considers both raw WOWY and adjusted WOWY can be used as evidence for Jordan having the GOAT prime/peak

Yes, a sample of 154 off games (that include 94 and 95!)

Yeah, so...are you sure about that? Ben doesn't explicitly tell us what years he includes and excludes but...

there is indirect evidence for a player when his teammates leave the lineup. Let’s say we wanted to know how much Scottie Pippen contributed to the Bulls +9 point-differential in the early ’90’s. In 1994, when Michael Jordan left the Bulls, we could infer something about Pippen based on the change caused by Jordan’s absence.

So while Bill Russell didn’t miss as much time as Jerry West, there’s a bevy of evidence about Russell left by his teammates and all of the time that they miss over the years.

Ben distinguishes between "indirect evidence" and WOWY and we know Ben isn't including the likes of 2011(Lebron), 1970(Russell). I'm not even sure 1995 qualifies as a "prime" year here. Equally concerining is the fact that the players you're comparing Jordan to like say, Lebron or Russell, have missed almost no time outside of those "indirect samples" which WOWYR isn't including. At best, you're using a much smaller sample, spreading it thin, and then throwing it in a comparison with other players where the sample size is truly 2016 Fisher-esque. At worst, we're going off less than a game a season. Either way, your sample is smaller. That part isn't really up for debate.

Re: Ben doesn't explicitly tell us what years he includes... yes he does! Pleases refer to the database links I cited in the previous comment

Re: does he include 154 off sample in WOWYR, that's a good question. I believe the rules he set are to consider off samples as games that a player misses for the teams they're on. For WOWYR, he doesn't use trades/moves to help buff up the off sample (though I wish he would, just to see how the scores differ). Jordan was on the team and playing in 95, so he's guaranteed to use the off sample in 95. Jordan was also still on the payroll and technically on the team in 94, so that's why I believe he used the 94 season as an off-sample (compared to someone like Russell in 1970 Celtics, who was no longer on the Celtics' payroll).

And all this says nothing about the fact that WOWYR sample is getting larger and getting stabler by incorporating the changes in teammates and opponents too.

Oh, but "this" was said, you just ignored it(I'm noticing a pattern):

For the sample I think you're using(1987-1997), Pippen only missed significant time(as a starter) in 1989(30ish games) and 1994(10). IOW. this "10 year adjustment" is mostly based on Pippen's exploits as a second-year player.. The "adjustment" being applied to Jordan's ben teammate isn't "correcting" the data, it's distorting it. Not only have we reduced our sample size by a factor of 10, but we're also treating Pippen's exploits as a second-year player as if they are relevant to what Pippen was doing in 1991. (Side-bar: You seem to be including games where rookie Pippen didn't start from 1988 while excluding games where Grant sat in your "without" here. Is there a reason for this?)

This is actually an issue Ben himself outlines(check those articles you linked), and probably why in his own impact evaluations, he mostly focuses on concentrated stretches, not "10-year samples" featuring about a 10th of the relevant evidence. Moreover. even after we've used a 10-year regression to turn a mountain of data into pebbles, Jordan is still well off the very best

So my sample has:
-666 on-games for Jordan,
-154 off-games for Jordan,
-555 off-games for Jordan's various teammates (479 if you exclude rookie Pippen, maybe off slightly if I miscounted Grant's game numbers?)
- ~thousands of data points for each of Jordan's teammates' on-games
- ~thousands of data points for each of Jordan's opponents' off/on-games

... while your sample has just
-277 on-games for Jordan
-291 off-games for Jordan

Your complaint about young Pippen underrating Jordan's teammates' value accounts for 30/479 = 6% of games used in Jordan's teammate adjustment, while having no effect on the large sample of Jordan's on games, off games, Jordan's opponents' on games, or Jordan's opponents' off games.

Re: "Moreover. even after we've used a 10-year regression to turn a mountain of data into pebbles, Jordan is still well off the very best"
Again, ... this just isn't true. Have you actually checked the links I sent?

From the official adjusted WOWY source:
Prime Jordan adjusted WOWY rank: 4th all time
Prime Russell adjusted WOWY rank: 30th all time
Prime Kareem adjusted WOWY rank: 29th all time
Prime LeBron adjusted WOWY rank: 8th all time

4th > 8th > 29th > 30th. Prime Jordan's adjusted WOWY > Prime Russell's, Prime Kareem's, Prime LeBron's. I don't know how to be any clearer

Part 2: RAPM Discussion

Moving on...

Again, the fact that the per-season off sample does not matter because that’s not how the calculation works. The sample for WOWYR is just bigger for both on and off games. And, better yet, it's during Jordan's prime during the time period we're interested in, unlike 1984 or 1986.

Honestly at this point, I shouldn't be suprised, but the focus wasn't "84 or 86" Jordan, it was 1988 Jordan with me very kindly assuming the Bulls didn't get any better after Jordan got drafted(cough Oakley cough). Put another way, I juiced Jordan's data, and he still fell short. And, yes, large 82 indirect samples are useful, at least according to the father of 10-year WOWYR:

If Jordan left and the team remained a +9 team, then it would be fairly safe to infer that Jordan was not the reason the Bulls were +9…which tells us that key remaining players on the team, like Pippen and Horace Grant, were the ones responsible for the large point differential.

Heck, it seems Ben preferred an "indirect" 20 game sample over that "10-year one" when it came to evaluating Lebron. If only he'd remembered to adjust!

Well, I've never had to convince someone why they should trust Adjusted Plus Minus more than raw plus minus before.

But 10-year WOWYR is not adjusted plus-minus. And as we've covered, actual RAPM doesn't favor Jordan(8th in a narrow field, best year is well off several from Lebron). Your analogy falls to pieces with a cursory understanding of what's being used.

Re: RAPM, as I've already said, the data we have fully consistent with Jordan being GOAT level (Top 4 all time) in RAPM.

There’s a problem with comparing Jordan vs LeBron’s RAPM straight-up: we have different samples for them! With LeBron, we have RAPM for every season. With Jordan, we’re missing many of his best years (no data for 89, 90, 92, 93), the seasons we do have are often incomplete, and the incomplete samples we have often under- or overestimate the Bulls. This under- or overestimation could be a change in Jordan’s teammates, but it could also be a change in Jordan himself.

For example: Jordan's RAPM in 1991 is calculated on the biggest underrepresenting sample in Squared2020's database. Across a full 57 games (i.e. not minuscule… as big of a sample as we're going to get), the Bulls' opponent adjusted win-pace was only 53 wins << their full-season 61 wins < their full-season-SRS-predicted 64 wins. That means that the Bulls (either Jordan or his teammates, or both) were dramatically underperforming in the 54 games we have. They underperformed by 13.39% compared to their true wins and by 17.45% compared to their SRS-predicted wins.

So how would we do an apples-to-apples RAPM comparison of Jordan vs LeBron? We would need to compare equivalent samples! That means:
1) Since we don’t have all Jordan’s seasons, we should only looking at the same amount of seasons for each (i.e. 5 prime-ish seasons each). But which LeBron seasons do we pick? We might compare LeBron vs Jordan at the same age, or at the same stage in their career… I’ll try both.
2) Since Jordan’s incomplete samples under or overrate him, we should adjust Jordan’s numbers up or down based on whether the incomplete samples under- or overestimate the Bulls. There’s some uncertainty involved here (e.g. if our sample underestimates the 91 Bulls compared to their true full-season value, we don’t know how much this was from Jordan playing worse vs his teammates. The fairest thing to do is consider everyone underperformed equally… but this assumption does increase our uncertainty. Jordan's true numbers could be better or worse).
So what are their RAPM scores?

Spoiler:

Jordan's RAPM scores:
1998 - 6.15
1997 - 5.84
1996 - 7.79 [sample size: 33 games]
[7.97 adjusted for their full season win pace.
7.75 adjusted for their full season SRS]
1991 - 6.4 [sample size: 57 games]
[7.34 adjusted for their full season win pace.
7.75 adjusted for their full season SRS]
1988 - 7.47. [sample size: 43 games]
[7.29 adjusted for their full season win pace.
7.58 adjusted for their full season SRS]

LeBron's RAPM scores:
2019 - 3.44
2018 - 1.56
2017 - 6.62
2016 - 8.62
2013 6.4
2012 - 9.29
2009 - 8.84

How do they compare looking at them at the same age? (1988 ~ 2009, 1991 ~ 2012, 1996 ~ 2017, 1997 ~ 2018, 1998 ~ 2019):
12 > 09 > 96 > 91 > 88 > 17 > 98 > 97 > 19 > 18
But LeBron was injured in 2019, so let’s compare their parallel seasons by career stage (to avoid using LeBron’s injury year).

How do they compare looking at seasons in the equivalent stage of their career? (88 ~ 09 as both are young athletic high-box-score breakout seasons; 91 ~ 13 as both are commonly considered their best individual years that blend young athleticism and old experience; 96/97/98 ~ 16/17/18 as both are older 3-year finals runs):
09 > 16 > 96 > 91 > 88 > 17 > 98 > 13 > 97 > 18

In both cases: LeBron has higher highs, but he’s also more inconsistent with lower lows. Interestingly, Jordan actually has better average RAPM in this more apples-to-apples 5-season comparison.


LeBron vs Jordan in 5-year RAPM:
LeBron average (choosing equivalent seasons by age): 5.95
LeBron average (choosing equivalent seasons by career stage): 6.41
Jordan average (adjusting smaller samples by full-season win pace): 6.92
Jordan average (adjusting smaller samples by full-season SRS): 7.01
So small advantage MJ when taking multi-year averages. Notably, Jordan’s advantage grows if we change these stats from per 100 possession to per game or per season. LeBron missed 10% of his games in the first sample (though that’s mostly from 2019, which is why I prefer the second sample). LeBron missed 5% of games and played 3% fewer possessions than Jordan in the second sample (for a 8% further reduction in total RAPM... which would be 5.89 using his better sample). Jordan missed 0 games in these seasons.

Jordan's advantage in possessions and fewer missed games also shrinks the gap in single-year RAPM: LeBron's best RAPM year when adjusted for his per-season rate is (9.29 RAPM/100 possessions)(71.1/100 possessions per game)*(62/66 games played) = +6.2 in 2012. Jordan's best is +6 in 1988 (and again we don't have 89/90, etc.). Seems pretty close to me. Now this single-year analysis allows us to cherrypick LeBron's best years to make him look better. But this seems slightly unfair, given we're missing many of Jordan's best years and thus can't cherrypick his samples. The fact that Jordan's right there in single-year samples and actually has an argument over LeBron in multi-year samples seems pretty complimentary to Jordan.

Side Note:

Spoiler:

We also have a sample for rookie 1985 Jordan and for old Wizards Jordan. I didn't include them, since those years seemed clearly a level down from ~prime 88-98 Jordan or ~prime 09-19 LeBron. But in case you're interested in the rookie/old-age comparison:

1985 Jordan: 5.03 [sample size: 32 games]
[4.96 adjusted for their full season win pace.
5.22 adjusted for their full season SRS]
2002 Jordan: 0.9
2003 Jordan: 1.46

2004 LeBron: 1.1
2006 LeBron: 3.75
2023 LeBron: TBD
2024 LeBron: TBD

So 21-year-old Rookie Jordan looks better in RAPM than LeBron when they were at the same age (21) and when they were at the same career stage (rookie year).
We don't yet have complete numbers for 38 or 39 year-old LeBron to compare to Wizards Jordan. But if I had to guess, I'd say they'd be better than Jordan's +0.9 or +1.46.

Source: Squared2020's historical RAPM for Jordan. The standard source for RAPM for LeBron (Goldstein).

It really does.

Unless you care about rs win totals than championships, it does not. Russell looks much better. Wilt scales. Jordan still loses. "Russell won 11 rings with a 35-win cast" is not a win for the guy who won half as much. It just isn't.

I claimed 10-year prime adjusted WOWYR has Jordan better than Russell.
You said "it just isn't."

Here's the official source for Prime 10-year adjusted WOWYR: https://thinkingbasketball.net/metrics/wowyr/
Jordan's +8.3
Russell's +5.9
The specific 10 year samples are chosen to be their best 10-year runs.

8.3 is indeed greater than 5.9. Jordan's 4th all time is indeed greater than LeBron's 8th all time, Kareem's 29th all time, and Russell's 30th all time (see aforementioned link and aforementioned Backpicks Top 40 articles). It just is.

There are real impact metrics based on actual impact data that portray prime Jordan as GOAT level over LeBron, Russell, or Kareem.

There are also "real impact metrics based on actual impact data" that potray Fisher as better than Lebron, 2012 KG as "GOAT-LVL", and Manu as better than Duncan. They key word here was "basically", because if one is determined enough, you can always take the Fisher approach and find tiny islands here and that rebel against the sea. I've specifically thrown in caveats like "generally speaking" and "most comparative frames", and "basically" for this reason. Though, in Jordan's case, even those islands only get you so far, as he still comes up short against Magic, D-Rob, and when you adjust for lower championship tresholds(and yes i'd say not making it is unreasonable if you value titles over regular season wins), he comes up well short of Russell and Wilt.

Like at this point you're neglecting basic addition to defend the samples you're using(I don't even know how you got "handful of games" in the first place actually), you're probably overshooting how much there is significantly(Ben consistently does not include "indirect" evidence in 10-year WOWYR), and even if you weren't, that would not change there's even less to go off for the players you're making a comparison with. Never mind, you trying to appeal to a non-existent consensus WOWYR's own creator probably doesn't support because you apparently don't understand the difference between lineup-level and game-level data?

It just isn't there. If you need to reach this hard just to place Jordan over "some" of the guys he's otherwise below, and you still can't get him to the top, it's probably not worth it.

[/quote] Yeah, I think I just disagree here for the reasons mentioned in this post and for the ~20 stats listed in the past few pages. But that's okay! We would get any of these fun debates if we always agreed all the time

[Lost92Bricks]

"GOAT candidate"? Only on LebronGM is this even questioned.