OhayoKD wrote:penbeast0 wrote:This is another interesting take and I like it a lot. I'd really like to see your player groupings that gave you the numbers. That would be a fascinating read.
Not my take unfortunately. Just me being a copy and paste merchant again
Maybe someone here could ask them though. I'd like to see the process too.
Cool stuff! No worries, there's definitely value to being a copy and paste merchant. I hadn't seen this before you posted!
They have (a little) more information y'all are asking for on their website:
-Crafted Plus Minus: basically some weighted average of some stats they like (Darko gets the most weighting, basketball index's Lebron stat, BPM, 3-year luck adjusted RAPM, and something called 'drip'):
https://craftednba.com/glossary . There's a bit of mixing of intents in these stats (Darko is meant to be predictive of future value; BPM is meant to be descriptive) and presumably some of the stats they chose were for accessibility (no EPM despite it being the best descriptive stat available; presumably using Basketball Reference BPM instead of Thinking Basketball BPM despite the latter being much more accurate, presumably because both are behind a paywall). Still, they do span box stats, pure impact metrics, and hybrid metrics, so you can think of this as some average performance across a bunch of different stats (although not always the best stats available).
-Player archetype:
https://craftednba.com/player-roles. They say player archetype on offense is done based on work from Nylon Calculus here
https://fansided.com/posts/5-nfl-general-managers-who-would-win-olympic-gold-medals-at-salary-cap-gymnastics, although it looks like they've cut two of the categories from the article (there's no longer categories for Domineering ball handlers and Tall ball handlers). They use a machine learning method called k-means clustering to group offensive players.
It looks like the link for assigning defensive roles is broken.
Some notable players across the categories:
Offensive:
1. Primary ball handlers: Luka, Shai, Brunson, Haliburton, Lillard, Trae, Harden, Ja
2. Secondary ball handlers: Tatum, Curry, Booker, LeBron, Kawhi, KD, Paul George, Jimmy Butler, Jamal Murray, Anthony Edwards
3. Versatile Bigs: Jokic, Embiid, Prozingis, AD,
Defensive:
1. Off Ball Defensive wings: Jared Vanderbilt, GPIII, Ausar Thompson, Matisse Thybulle, Chris Paul,
2. On ball stoppers: Caruso, Kawhi, Jalen Suggs, OG Ananoby, Derrick White, Giannis, Paul George, Butler, Marcus Smart, Jrue, Edwards, Tatum, LeBron, Siakam, Wiggins)
3. Drop bigs: Gobert, Green, Adebayo, Davis, Jokic, Wembanyama, Embiid, Brook Lopez, Chet, Jarett Allen, Porzingis)
Overall, it's interesting stuff and I'd love for there to be more research into this kind of thing!
What's valued?
The top 2 archetypes are perimeter players, and 4/5 of the top archetypes are perimeter players. This suggests on average, at least in the current NBA, perimeter players are generally more valuable than bigs, and that the value of big man defense (plus positive offense) is not enough to overcome the value of perimeter offense (plus positive defense).
What's not valued?
Spot up players are ranked as the easiest to replace. Does this mean shooting isn't that valuable? Maybe, but maybe not. I wonder how much this is unique to this era, when shooting is far more ubiquitous. The spotup categories may also filter against star players (e.g. all the best star shooters, e.g. Curry, Durant, Lillard, KAT/Jokic et al. are seen as clearly not in the spot-up category, despite being such strong shooters)... so this may just suggest that the best shooters get assigned to different categories, more than saying the low ranking of spotup archetypes is an indictment of the value of shooting.
Possible future work: Research into synergies (e.g. 2/3-player RAPMs vs archetype pairs) would be especially interesting, as would research into how these archetype rankings change as teams get better or worse.
Some limitations:
-Of course there's limitations in the stats used, like I mentioned above.
-The k-means clustering may do well in aggregate, and may be one of the best things possible with current stats, but...
a. people might disagree on specific cases (e.g. are Curry and LeBron really in the same offensive archetype?),
b. the names of categories may not match what people think of (are GPIII, Ausar Thompson, Thybulle, and Chris Paul really defensive 'wings'?), and
c. there's fewer defensive stats so defensive categorization may be more difficult (are Draymond and Davis really a drop big? are LeBron and Giannis really on-ball stoppers more than off-ball wings?).
Having players assigned to the 'wrong' archetype (at least as we would think of it) would of course also increase error in the estimated value of each archetype.
-Some categories have small samples so the median may be noisy (only 8 pick and pop bigs, at least in 2024... do we know if this research is for the current NBA or over a range of time?)
-Median value may not match the upper standard deviation or the ceiling, and many of the best players break traditional categorization in some ways, so there's limits into how much we can apply this into players at the top of the league.