f4p wrote:You seem to have more faith in less predictive AND flexible metrics on the basis that... Curry being the best regular season player a bunch is absurd.
i think they all are probably not super predictive, writ large, across the entire league. i suspect on/off stuff that you like is going to win out on the whole in prediction simply because it is literally using the score from the game (based on lineups) to be predictive.
...I don't think you're reading very carefully here
A greater test of metric efficacy is observing how well metric values predict team net rating when rosters change. Roster continuity in and of itself is predictive of team ratings, meaning teams that have little roster turnover from one year to the next tend to perform better. This is likely due to good teams keeping good players with secondary effects from team chemistry and club strength (coaching, player development, training staff, management, etc.), but a good player metric minimizes any team effect that could be present. So how much is a particular metric dependent on teams sticking together?
The metrics that directly draw from winning(EPM, RPM) do the best in terms of stability (or as i called it, "flexibility") AND predictivity. PER finishes dead-last on both fronts. As of now, your theory isn't really supported by anything. In fact, everything we have on this contradicts your supposition. Player-tracked data that draws from plus-minus grades as more stable than data that draws heavily on box-stuff. Looking at stuff without player-tracking, RAPM is more stable predictive than box-aggregates like PER and WS/48(keep in mind player tracking doesn't exist for stuff like BPM or RAPTOR pre-1997). Simply put, "Impact on winning" is doing a better job isolating and identifying the pieces of the vase than "numbers". Why? Because they do a much better job capturing defensive impact. Speaking of which...
but my read of what seems to end up with people winning rings and playoff series is that the box metrics are not guiding us that wrongly.
The most winningest player in nba history grading out as a role player isn't a red-flag for you? How is this a lesser issue than RPM sometimes disagreeing with you on RS Curry?(You're confusing some stuff there too, but more on that later). We don't need to talk in vague abstractions. The reason Jordan does disproportionately well in box-heavy metrics is because box-heavy stuff is looking at his steals and blocks per game and concluding is he's one of the best defenders in history(DWS actually matches Hakeem's in 88). Impact stuff looks at how MJ's presence correlates with the Bulls' defense getting better or worse and concludes that Jordan isn't on par(or paticularly close) to even second tier defenders like Kareem, Kawhi, Lebron, and Pippen. Much has been written on why we should expect approach two to be more effective and from the studies that have been done, it seems approach two leads to more predictive AND stable data.
On that note, If you're going to use "history" as a check, you should probably look at whether championships or winning(tho this is arguably just a more crude variation of WOWY) supports PER being so much lower on primary paint protectors than plus-minus/impact/WOWY is. I'd guess if you really looked at this, the answer you'd find is no. Lebron and (to a lesser degree)Kareem are able to survive getting their defensive contributions diminished because they are also goated(or within range) on offense(as well as defense making a smaller portion of their value).
Your 2018 complaint for Curry is also off. RPM says curry was the best per possession. When we switch to wins added(thus looking at the totality of contribution), curry ranks 10th. Also...
and kyle lowry leading the league in RPM is probably not something anyone is going to be able to support either.
Why do I need to? My claim wasn't "impact metrics are perfect", it was they're more useful than box-stuff. How does "but kyle lowry" address the much larger sample that has been pooled multiple times indicating that the metrics you prefer are even funkier? You're not providing support for your position, you're not addressing the theoretical defensive concerns, and you're not applying scrutiny evenly.
Why are you comparing hakeem or duncan to pj tucker when assessing if they're underrated? Why is "does rpm's assessments of players always agree with mine?" enough to throw out plus-minus based analysis but the most winningest player ever(on, from what we have, at least some not-spectacular help) looking like a scrub not make you reconsider?
I'm not saying your theory is impossible, but you're using it as if its been proven(you aren't even considering the defensive side of things) when its basically just unsupported conjecture (at least right now). Can't we agree, at least for now, the available data isn't favorable for your theory, and you probably should at least consider the other stuff? I'm guessing you'd agree steals+blocks =/ defense(and as I highlighted earlier, smaller players will often rack those up thanks to the "anti-gravity" offered by bigger defenders). If bigger players(primary paint protectors) are consistently looking better relative to Jordan(and other steal-racking guards like curry and kobe) in all the non-box stuff, and worse in all the box-stuff, isn't that notable?
Seems at least plausible to me that's why the metrics you are pushing as stable are grading dead-last in stability, while the metrics you're expecting to be the least stable are grading as the most stable, and why WOWY(where we can draw the largest and most inclusive possible samples) is painting a very different picture than the one you're extrapolating from PER-esque stuff(2-way race with lebron for GOAT vs a top 10-er in the rs with playoff elevation).
Like, seriously dude. I am giving MJ the assumption that his team did not improve whatsoever after he was drafted(contradicted by their defensive improvement and collapse with oakley coming and going), and the absolute best you get here is 23 wins in 1988(lower if we use srs). If we took PER/Box-only BPM/WS48 at face value here, then the Bulls magically turned into a 10 win team after winning 27(in line with 82 game srs) in 1984.
Even RAPM is somewhat artificial. Stuff like PER is even more artificial. At some point things need to be at least somewhat in line with observable reality. Priors are fine as a starting point, but dismissing everything that might challenge that prior on the basis of unsupported conjecture is a recipe for confirmation bias.
you could play the jordan simulation 100 times and i don't think you are getting the 2011 finals. that's what i mean by "steady". nothing in his distribution of performances tells us that was within a reasonable number of standard deviations to actually happen. nothing tells us jordan was dipping low enough to reach the lows of some other guys, even if he "only" played about 70% of the playoff games of the very longest careers (and more games and years than some other top 10 guys).
But we don't need to play a simulation? In 1993, Jordan was arguably saved by help from a big potential choke to a less talented underdog. In 94 Jordan stopped playing basketball partially, at least by his own admission, because he needed to recuperate emotionally from the last 3 seasons. In 95 he posted a worse season than Lebron's 2011, and in 96 he was saved by his cast, at least potentially, from what would have been the greatest collapse in nba history. Again this "reliable" angle only really works if you find reason x, y, or z to dismiss all the evidence of "unreliability".
This is the problem with going off "expectation" as opposed to "help". Expectations for a player often are determined by that player himself.
Jordan is an underdog in 1989 vs the Cavs while the cavs are a favorite vs the Raptors in 2018. But did the Bulls actually have less help than the cavs? Is beating a playing-through injury Price with a buzzer-beater a better achievement than sweeping the similarly srs'd raptors? Okay 95 is not "prime MJ" because MJ chose to retire(and he had fair reasons), but the "prime" period here is not nearly as long as it is for the "less reliable" primes of the players you're comparing him to.
You're arbitrarily lowering the bar. Compare apples to apples, and "reliability" stops being a winning case.
i like being challenged. but i suspect we all agree more with stats that reinforce our priors. i don't think it's a shock that the steph fan loves the EPM's and RPM's of the world.
Maybe, but i really prefer to assess people on what they're doing as opposed to what might motivate them. If I'm a steph fan because I think RPM is >>> PER(which gives steph a best rs case in 2016), am I a steph hater when I advocate WOWY use in certain contexts over RAPM despite WOWY potraying Curry alot less favorably relative to his all-time peers?
Everyone is capable of bias, but that's not a good excuse to rely on circular reasoning. And I imagine if you tried to do a more rigorous analysis of my posts(kind of what I tried to do with NMR earlier), you may find your priors regarding me are off
The biases driving these arguments might be a bit broader than you're imagining. And fixating on potential bias as opposed to the substance present in what people are saying can lead to misunderstandings.
Case in point...
...but it starting to feel like you're looking for reasons to keep using stuff like PER(which game-score is a unadjusted extension of). All the articles doing this comp have those metrics grading out as the worst.
i like that the box metrics have a steadiness to them from season to season.
...they don't. At least via these studies. And also...
i like that we have them for basically all of nba history instead of some stats where we go back like 10 years and could never use them to compare to older players.
...their utility in historic comps is greatly limited by the lack of complete data(I mentioned this before btw).
WOWY type stuff is like the only real evidence we have for players pre 1980(unless you want to go off 2.2/szn wowyr samples) which isn't subject to blatant inaccuracy(and WOWY seems to be matched by plus-minus much, much better than its matched by Box-stuff).
but a 29 vs 24 is going to give me a lot of confidence i'm looking at different qualities of players.
31 vs 26 is a similar gap. Are you confident that 2016 Curry is a different quality of player than 2003 Duncan?(who won 60+ games with the famously stacked early aught San Antonio). How about Hakeem(26 PER, 23 at his perceived peak) and Bill Russell. Then again, 2016 Curry's defense was obviously on par with Mark Eaton's(DWS) so I guess I'd be extremely confident too.
i think they aren't nearly as broken as you think they are for guys who actually fill up the box score.
Yeah sorry, but MJ's defense = Hakeem and Curry's defense = Eaton coz steals doesn't really fill me with confidence. Even if, and right now there's not much reason to think this is true, box-stuff was actually preferrable to winning stuff, I'd think it would be wise to apply caution when they tell you guards are a "different quality of player" than players who can actually lead elite defenses. Like Is this really what "90% to truth" looks like? And let's talk about what this is "based on" here...
PER, WS48, and BPM and i feel like i usually get about 90% of the way to the truth based on everything that seems to have won games over NBA history.
It's hard to imagine this not looking like a crude variant of WOWY...which correlates far more strongly with plus-minus stuff(actually potrays jordan less flatteringly than aupm or on/off do).
Maybe you could elaborate on what you are using to assess "seems to have won games over nba history?"
hell, at one point PER and WS48 were state of the art and no one wanted to listen to them because we had points and rebounds to look at and who knew what these "advanced" stats were anyway (advanced apparently as in they involved a tiny bit of math).
Well a little-bit of math and entirely arbitrary weightings for different subjectively defined and recorded game-actions with a perceived correlation with winning(steals can be positive and negative indicators but the box-score does not record blown coverage so...). The relationship to winning(assuming we agree helping teams win is the main objective of basketball players is actually much more convoluted with this metric, than the ones you prefer it to, which may explain why it doesn't seem to grade so well when tested a bit more rigorously (as opposed to taking a couple of data points and comparing it with one's own priors).
Using the "history of winning" to justify your preference for metrics that artificially create a ocean between their methodology and "winning" seems off.
i'm insulting the strat more from the perspective of it being weak sauce. top 20 all-time guy just takes a year off. hey, things have been so great with our talent and infinite payroll, we can just chunk a season away. peasants like damian lillard, james harden, lebron james, and everyone else not in a perfect situation who actually try in the regular season are crazy for not just relaxing like we do when it looks like we'll only be a 7th or 8th seed. who wants a tough playoff struggle and possible first or second round loss when you can just get a high draft pick? silly gooses. we'll be back in 2 years to win and pat ourselves on the back for all the things that we've overcome.
Optimizing your chances to win a championship isn't weaksauce imo. Losing three of your key piences over a single off-season with your two centerpieces on the wrong-side of 30 is alot to overcome. Sometimes swallowing your pride and adapting to reality is better than trying to prove you're "tough".
yes, if he stepped out of a time machine, he would be like an alien and unstoppable. until rick mahorn got tired of it and forearm shivered him. but several weeks later when steph recovered, he would go back to dominating. but that's not really a fair argument i don't think. if he grew up back then,
Is "fair" the goal here? I don't really think generating a completely different player is in the spirit of the exercise and when you go with "growing up" you've massively increased the range of uncertainty. People are results of the context that creates them. This new player you invented isn't steph curry, and if we're being honest, chances are he(as with any player who grows up ina different time)isn't playing professional basketball.
Regardless, I'm going to put a pin in the Curry/Giannis stuff, because I anticipate progress there will be dependent on some sort of shift from one of us towards the other regarding the statistical stuff. MJ is just a different caliber of player depending on the methodology used, so I think its worthwhile breaking down that as opposed to talking past each other because we're dancing around the primary source of disagreement(for us and many posters on the board).
).
), but in the same thought also using Ben’s overall metric when talking about Jordan’s peak, perhaps as an effort to show a bit of equanimity to the idea of Jordan having the top peak — made for a good illustration of what I was trying to say.
)
