Djoker wrote:OhayoKD wrote:
They're the same type of data. Humans choose what to count and then put weights on what they've counted, That decades were spent enshrining a narrow set of approaches as objectively valuable does not magically give the formulas and inputs you prefer inherent value and pretending it does would get you discredited in any space with an ounce of serious academic rigor.
Beyond the extent you can justify the approach or weightings vs approaches/weightings that favor alternative players, your formulas are not legitimate evidence.
IBM of course is not a few games, Lebronny's tracking covers multiple years of full playoff runs. If sample size is the issue, then the solution is to increase the sample, not keep reinforcing a set of priors that have never been seriously tested because they produce outputs you find convenient.
I guess I see your point. However, it's still a fact that we have huge samples of BBR data and tiny insignificant samples of this other data. So don't blame me if I choose the samples that are bigger by a factor of a 100.
In an absolute vacuum of alternative information? Sure. But I would say when possible, one should be making these choices largely based on what the data not biased by these human choices suggests. And this is the main issue here. Magic's non-box outpaces Jordan atm, yet PER-esque approaches say Jordan was way better. Even if you find it absurd, a formula or all-in-one which put almost all it's weight on assists and rebounds would probably line-up better with the cold data here. Why assume the Per-approach is more useful than the latter when comparing Magic and Jordan when the latter gets closer to the results? In general I think the box-score, like any eyetest/granular info, must be supplanted with results to have real value(presuming we are trying to measure impact or corp or whatever). For this specific comparison the approach doesn't lineup with the data, and the sample being 100 or 1000 times larger doesn't really change that. When priors fail to explain a phenomenon, one should shift the priors they are using.
Also worth asking is.. How do you figure out which data is useful and which isn't? Why is tracking creation the way you do it more valuable than tracking say box creation which Ben Taylor does?
I will cover three things here. Lebronny's stuff, Falco's tracking, and my own tracking.
only tracking Ben has done there is for box-creation is opportunities created which he never applied to the 80's or 90's. Box-creation readjusts a bunch of box-score stuff to try to regress to OC.
Lebronny's approach is pretty much the same as the tracking Ben was trying to regress to, the main difference is he actually has posted it for
several years worth of full playoff runs while Ben has not, and we actually have it for the 80's/90's (we don't for Ben). Both have an issue with peer-review currently, but unlike ben, Lebronnygoat has actually provided the games for vetting in a drive-link, and is planning to provide time-stamps.
The primary advantage over lebronny and ben's oc over assists is
it includes plays which didn't end up being converted. That to me is a massive advantage since obviously how many of your creations end up as scores is largely a product of team-circumstance.
Falco's approach makes an attempt at not just tracking all types of creation, but also differentiates between
quality with 4 levels accounted for. They actually account for creation which is not rewarded with an assist (Box-creation does not), and unlike either box-creation or oc or lebrony's tracking,
he's provided time-stamps and footag meaning his **** can be directly peer-reviewed and vetting. The main disadvantage is because of how comprhensive and detailed his tracking is(he was simulteously doing the same thing with defene and rebounding and scoring and all sorts of information simply not covered by any existing approach), it's extremely time consuming and will probably need alot more helpers to produce at scale. It has also been vetted by multiple people including posters here. The quality stuff is also more subjective, even though he's offered definitions, and he hasn't counted anything there.
My approach is basically addresses that by counting something which can serve as a proxy for creation quality regardless of whether you agree with my qualitative judgements. That said with my tracking there are two variants.
Full-Game tracking
Assist-tracking.
With Full-game i'm basically trying to do what falco did but also put some countable numbers people can compare. It carries the advantage of looking at non-assists, having time-stamps and specific stretches of footage that can actually be peer reviewed, and offering something material to compare.
The thing that there is more of at this point is assist-tracking and here you lost the advantage of non-assists being tracked.
Here is the case for at least incorporating the others in analysis even with Box-Creation already there as a model.
1. General utility - We know from tracking data the more outnumbered a defense is, the more likely an offense is to score. Stands to reason the more defenders you take out, the likelier your team is to score. Box Creation flatly does not account for that. OC didn't either beyond assuming that moving an extra defender is valuable. Assist Quality is simply not something tracked in BBR or OC
2. Localized utility - We also know Jordan's conventional all-in-ones outputs, do not lineup with his impact in comparison to by reputation (and by all the types of tracking listed above) top-tier passers and more ball-dominant players like Nash, Magic, and Lebron. You yourself have argued that when Jordan is averaging triple doubles, he's providing more playmaking than Lebron in periods he provides less assists. Yet, point-Jordan did not come anywhere close to Lebron in demonstrated impact and the gap widened when we compared him to point Lebron. Magic should be absolutely cooked by Jordan if you really buy his assist averages, yet his impact data is actually a little bit better in general. This is a phenomenon that warrants explanation and I think we get one with Jordan getting outpaced by those three in pretty much all the above approaches.
The reason for using the defensive stuff, paticularly rim-load I think is even clearer. A minute fraction of defensive plays are currently counted by the box-score
and there is a big dissonance between Jordan's defensive data and what these outputs say about him defensively.
I just don't see much point in ignoring alternative stuff when it lines up with the actual data much better than the conventional stuff does for this comparison.
Another perk of all these approaches is they don't include cooked numbers that bolstered 1988 MJ's bbr more than likely anyone else in the league.
On/off is even in the regular-season and then it collapses for Jordan. And Magic is clearly advantaged in WOWY (something you cleverly side-stepped by throwing in games Magic did not play with his "with")
Selective context once again. You know team was even more injured than Detroit allowing the Bulls to punch well above their weight statistically?
https://www.basketball-reference.com/playoffs/1988-nba-eastern-conference-first-round-cavaliers-vs-bulls.htmlEeking out wins against injured iterations of 80s Lebronto is overrated imho.
Overall? No. Compared to the Lakers
with Magic? The Lakers posted a net-rating of
+7.2 more than
doubling Chicago's
3.5 with Jordan. The reason they weren't "much better" is because they collapsed to
-3.8 without Magic in the 10 games he missed. A convenient data-point to ignore when arguing Magic had "infinitely more talent" I guess.
Even with the PS data, Jordan's ON-OFF is still in the same ballpark as Magic.
If you insist on using 10 games without Magic (which is a small sample to draw any major conclusions) then you also have to penalize him for missing 10 games.
Anyways here are large WOWY samples for both guys. Works out well because it's 8 years and they had good teams around them.
1984-1991 Magic
With: 454-149 W-L, +7.42 MOV, 60 PW
Without: 29-24 W-L, +0.23 MOV, 42 PW
1991-1998 Jordan
With: 400-103 W-L, +9.38 MOV, 64 PW
Without: 90-63 W-L, +3.38 MOV, 48 PW
Magic adds 18 PW and Jordan adds 16 PW. I actually find Jordan's lift more impressive considering they are considerably better with him in the lineup. +9.38 MOV and 64 PW over a sample that huge is incredible.
Meh. Magic still has more than a full point of MOV-differential advantage here and the time-stretch excludes the season in question for Jordan. It also seems wierd to me to favor samples which don't include 88 MJ for the 1988 POY. It's also worth noting on a per-season basis that's not a bigger sample than 1988 Magic's off.
Ben's studies show that missing 10 games has a neglible effect on title probability and...despite those missed 10 games hurting the Lakers regular-season seeding...The Lakers still won anyway.
It also so happens 1988 Magic Johnson is one of those few examples of a player winning with a team that was very bad without him. As someone who thinks context should be applied consistently, not simply when it suits one's prior, I will note that the Lakers missed Micheal Cooper(the guy Jordan called a fraud for not stat-padding defensively like he did). Every other cog of their rotation played all 10 games rendering it a uniquely clean sample as well as one of the most impressive signals of the era(more impressive than any of Jordan's at any rate).
I'd say "dissapointing" is the right word there. That expectations were so low that replicating the results two of his less acclaimed contemporaries was deemed near-impossible is as much of an indictment on Jordan as it is on his teammates.
Bad without him based on a 10-game sample. Again you're looking too much into a small sample. A larger 53-game sample over 8 seasons from 1984-1991 posted above suggests that the Lakers were respectable and definitely a playoff team without Magic. Which kind of makes sense considering the supporting cast is very talented.
If you're not willing to admit that Jordan's supporting cast in 1988 was much worse than Magic's, then that's not a rabbit hole I'm willing to go down into because that to me should be obvious.
The stuff showing Magic's team as respectable also still suggests 88 Mj was not as valuable. I don't understand what you're trying to argue here. We know the 88 Lakers are significantly worse than the average lakers team and we know they're an order of magnitude better with Magic than the Bulls are with Jordan. Jordan can have a significantly worse team and still not be as impactful...which is exactly what the most relevant "larger" sample suggests.
That an extremely clean sample from the year in question suggests the talent level is being overrated for that specific is also not something you just ignore if you're operating in good faith. Weigh it less if you want but the non-human informed data clearly favors 88 Magic over 88 Jordan in terms of general impact.
Magic won the title and has a monopoly on unbiased data for the year in question and the 80's in general. That seems like a pretty clear cut POY to me.
