DraymondGold wrote:Hakeem in Impact Metrics
Given all the discussion on Hakeem, I thought it might be interesting to go through all the available impact data we have. All of the stats we have are imperfect (RAPM has small samples, WOWY-based stuff is noisy), but I'd argue that they can still help us get a handle on a player when examined on the whole, in conjunction with contextual and film analysis. I'm going to start this post by summarizing the available 'pure' impact metrics, with the hope of getting more into the box/hybrid metrics, context, and film analysis in the upcoming days/threads.
So: how do available (pure) impact metrics rate Hakeem relative in the upcoming tiers? I'll use the previous projects' remaining Top 14 players plus Curry for these tiers. In chronological order: Russel, Wilt, Oscar, West, Bird, Magic, Hakeem, Shaq, Duncan, Garnett, Kobe, Curry.
Raw WOWY: This is probably Hakeem's most favorable impact stat. How does he look?
-10-year Prime WOWY: Oscar, West, Bird, Shaq, Garnett, Curry > Hakeem. (Magic barely behind).
-Multi-season lineup changes (the OhayoKD special): Russell looks GOAT-level and definitely gets above Hakeem. Others may look better as well, but there's less of a single database to check for these full-season trade/injury/rookie/retirement-based WOWY data.
Overall in raw WOWY, Hakeem only has a case over Wilt, Magic, Duncan, and Kobe. Russell, Shaq, and KG are better than Hakeem, as are all the all-time non-bigs.
Adjusted WOWY: if we adjust for teammates (in the same way you can adjust raw plus minus to make APM/RAPM):
-10-year Prime WOWYR: Russell, Wilt, Oscar, West, Magic, Shaq, Garnett, Kobe > Hakeem (barely Duncan barely behind, Bird behind; no Curry data)
-10-year Prime GPM (alternate calculation method for WOWYR): Russell, Oscar, West, Bird, Magic, Shaq, Duncan, Garnett, Kobe > Hakeem (Wilt behind; no Curry data).
-10-year Average between adjusted-WOWY stats: Russell, Oscar, West, Magic, Shaq, Duncan, Garnett > Hakeem = Kobe (barely over Bird or Wilt; no Curry data).
So in adjusted WOWY stats, Hakeem doesn't really have a case over the same bigs (Russell, Shaq, Garnett) plus Duncan, and all the all-time guards (Oscar, West, Magic, and likely Curry given Curry’s GOAT raw WOWY stats). He has a weak case over Kobe, some case over Wilt and Bird.
RAPM: We have small samples of Hakeem's RAPM, thanks to Squared2020. We have ~136 prime games (14 games in 1988 + 25 games in 1991 + 19 games in 1996 = 58 games in his 10-year prime, plus full-season data in 1997). We also have full post-prime and 9 games from his rookie season. Small samples can be very noisy (so larger uncertainty range), but 25 games in 1991 is big enough to not be entirely noise (particularly when boosted by the context of data from 3 other prime years, and data from 6 non-prime years).
How does Hakeem look in prime RAPM? His values are +1.82 in 1985, +1.52 in 1988, +3.19 in 1991, +3.50 in 1996, +3.37 in 1997. In other words...
Bird, Magic, Shaq, Duncan, Garnett, Kobe, Curry >> Hakeem (with no data for Russell, Wilt, Oscar, West). I.e., Hakeem's RAPM data is significantly lower than all the available players in this tier. But we're dealing with small samples, where Hakeem's teams underperformed vs their full-season rate.
What if we curve Hakeem's numbers up, based on his team's full-season play (so if Hakeem's teams performed 12% worse in the games we have vs their full-season rating, what if we assume the underperformance is equally from Hakeem and his teammates and so boost Hakeem's numbers by 12%)?
Hakeem ends up having +1.7 in 1988, +3.4 in 1991, +4.8 in 1996. Which is an improvement!... that still isn't enough to get Hakeem over the better years of literally any of the other available players in this tier.
Okay, if we still think the measurement is too noisy, what if we only compare the full-season data at equivalent ages (so age 34+). This gives a handle on how players aged, and maybe can help us infer prime value based on the decline.
Hakeem (age 34+): 3.37, 3.11, 2.62, 1.56, 0.5, 1.04
Shaq (age 34+): 1.97, 2.96, 0.62, -1.32, 0.43.
Duncan (age 34+): 3.26, 5.1, 5.24, 4.03, 3.04
Garnett (age 34+): 5.73, 6.89, 6.3, 3.46, 1.53
Kobe (age 34+): 0.74, 1.89, 0.18, -0.86
Finally, at least he's not last again!
So in (limited) prime RAPM, Hakeem looks worse than every available player in this tier. As an older player, Hakeem looks better than Shaq and Kobe (but under Duncan and Garnett). But again, this data is not ideal, and it may not capture his playoff improvement (which I'll try to get to in the coming posts/days).
Overall Takeaways: I’ll leave the box stats and playoff stats for a future post. These are important factors to consider (especially playoff stats for Hakeem), and I don’t want to rush them for now.
But in the mean time, these are all the true/“pure” impact stats we have for Hakeem. They all have limitations. Raw WOWY and WOWYR are both quite noisy with large uncertainty. The prime RAPM data use small samples, and are thus noisy. And the post-prime RAPM data is less noisy, but misses the seasons we’re actually interested in. However, taken together, they can still be used to help pin down the value of prime Hakeem. And what I've checked (so far) puts Hakeem towards the bottom of this tier.
Looking at players who are currently up for being voted in:
-Russell > Hakeem. The raw WOWY (multi-season lineup changes) data clearly favor Russell, as do all the adjusted WOWY data. We have no Russell RAPM data.
-Shaq >! Hakeem. Shaq’s better in every raw WOWY and adjusted WOWY stat, and his prime RAPM is significantly ahead. The only advantage Hakeem seems to have is age 34+ RAPM, so you’d need a very heavy longevity weighting to prefer Hakeem to Shaq.
-Duncan >~ Hakeem. You can argue Hakeem if you heavily weight raw WOWY data, but the adjusted WOWY data (WOWYR, GPM, etc.) favor Duncan, the available prime RAPM data significantly favors Duncan, as does the late-career full-season RAPM.
-Hakeem > Wilt. Wilt is a more interesting discussion. WOWY-based data has never been quite as high on Wilt. Hakeem is higher in the raw WOWY data, and is higher in the adjusted WOWY data (though not by much, well within the bounds of uncertainty). We have no Wilt RAPM, so it seems like the available data favors Hakeem.
For those who'd like to make a pro-Wilt argument, some of this can be explained with context: Wilt has down years in 1965/69/70/73, which are the very same years he’s switching teams/injured/retiring, so his available WOWY samples may be dominated by the down years. Alternatively, an argument for Wilt may focus more on box stats, or focus more on evaluating Wilt’s “talent” over his per-season “impact”.
As for the other players, WOWY/available RAPM data pretty clearly favors the all-time guards (Oscar, West, Magic, Curry) as well as Garnett and possibly Bird, so it may be time to start nominating them. A longevity-heavy, playoff-heavy weighting might be able to push Hakeem past some of them, but many of these players are favored in all the available impact metrics.
Since Hakeem has been nominated already (and the others haven't), this to me suggests that people are either not valuing the impact data we have that heavily (at least compared to box stats or film or qualitative analysis), or perhaps are valuing longevity and inferred playoff improvement enough to push out the non-bigs (despite their per-season advantage over some of these bigs). I'd love to hear thoughts on this.
Me personally, I'm not ready to have Hakeem at the very bottom of these tiers (particularly given the longevity, and possibly given the playoff-improvement pending more film/data analysis)... but I do find it somewhat concerning (for a Top 10 player) that his adjusted WOWY data is so low, and that he has literally no single (available) RAPM sample that would put him at strong-MVP or all time. I might have him closer to ~10th, rather than fighting for 4th.
Sources:
-Thinking Basketball's Prime WOWY/WOWYR dataset (the traditional source for WOWY/WOWYR)
-Curry raw WOWY was approximately calculated by me in the RealGM Greatest Peaks Project
-Squared2020's RAPM for historical players (the traditional source for historical RAPM)
-Goldstein's RAPM for post-1997 (the traditional source for RAPM)
Hakeem is ahead in Scaled WOWYRand Alt Scaled. He falls behind Duncan in 10-year Scaled GPM. Hakeem's Prime WOWYR is the same as Duncan's, and he has a higher Prime WOWYR. Hakeem falls slightly behind Duncan in average due to to the decent discrepancy between between their GPM values. However, their averages are nearly equal and Hakeem actually looks better in multiple variations of WOWY.