The Multi-Year WOWY Database

DraymondGold · Post #1 » by **DraymondGold** » Mon Jul 24, 2023 10:44 pm

~The Multi-Year Large-Sample WOWY Database~

There has been much talk about WOWY in the recent 2023 Top 100 Project, so I’ve decided to start a database for multi-year, large sample raw WOWY. But first, a bit of background.

What is WOWY?

Spoiler:

WOWY, or “with or without you”, is a raw impact stat like on/off. But where on/off looks at the difference between possessions you’re on the court vs off the court within individual games, WOWY looks at games. It compares how a player’s team performs in games they play vs games they do not play. Traditionally this is measured using SRS (margin of victory, or MoV, adjusted for opponents) for accuracy, though you could also calculate a WOWY score using Margin of Victory.
Pros: 1) WOWY gets at wholistic value. It attempts to measure a player’s overall impact on a game. 2) Since it’s an impact stat, it may be better than box stats at measuring subtler non-box forms of impact. This might include rim deterrence and positioning on defense, off-ball motion on offense, or even on-court communication and leadership.  
Cons: 1) WOWY measures situational value. Just like raw on/off can be biased by lineups, raw WOWY can be biased by rosters. Certain rosters may be better equipped to replace a certain player’s role when they’re gone, while others may not have viable replacements. Certain coaches may be better at adjusting gameplans without a star player than others, which may inflate or deflate a WOWY rating. 2) WOWY is also context blind. Like raw on/off, raw WOWY does not adjust for teammates. If Player A is in for part of the season then out for the next, while their teammates are improving over time, this might dent Player A’s WOWY, because raw WOWY doesn’t know to adjust for the fact that Player A’s teammate are changing (or even that other players are coming in/out of the roster). Likewise, if a player is out for a few weeks then returns to games while still injured, this will lower their raw WOWY, because WOWY doesn’t know the player is injured. Likewise, 3) WOWY is super noisy. In other words, there’s a lot of uncertainty. Even full career averages can have large uncertainty ranges: even if Player A has +6 WOWY and Player B has +5 WOWY, they may not be far enough apart for us to say Player A is more valuable beyond our range of uncertainty. 4) Relatedly, WOWY often comes in much smaller sample sizes than on/off. In smaller samples, we expect this data to have greater uncertainty.

What types of WOWY data is there?

Spoiler:

So what’s new with this WOWY data?
Where Thinking Basketball’s database only includes WOWY data from within a single season, here I’m calculating multi-year changes. This multi-year samples include changes to team performance over 1) rookie years, 2) mid-season/off-season trades, 3) long-term injuries, and 4) retirement years. This data has the benefit of being larger sample (helping Cons 3 and 4). However, since it’s over longer timescales, it’s more likely to be biased by other changes to the roster, improvement or drop-off from the player’s teammates, changes to the coaching staff or the league context, etc. (worsening Cons 1 and 2).
-Specific thresholds: I'm requiring at least 30 ‘with’ games and 30 ‘without’ games for a single team within a two-year span. It’s a somewhat arbitrary threshold, but the goal of this database is to have a sufficient sample size that there’s noise is minimized (as much as possible, although WOWY will certainly still have noise). I will use SRS for full-season data (e.g. for trades/injuries in the off season), MoV for mid-season data (e.g. for trades/injuries that occur mid-season). When relevant, I will include an ‘Alternate Value’ (e.g. if there are simple corrections for other roster changes, if you might use a different neighboring year, etc.).
-Sources: Basketball Reference for full-season SRS, Statmuse for mid-season MoV.
-The Players: Roughly the top 15 players. This isn’t meant to be a definitive list of the best WOWY scores or the true Top 15, just a list of the usual suspects that people are most frequently interested in. Those 15 players are… Bill Russell, Wilt Chamberlain, Oscar Robertson, Jerry West, Kareem Abdul-Jabbar, Larry Bird, Magic Johnson, Michael Jordan, Hakeem Olajuwon, Shaquille O’Neal, Tim Duncan, Kevin Garnett, Kobe Bryant, LeBron James, Steph Curry.

Future updates
This database is a work in progress, and may be expanded in the future. In particular:
-SRS: I used SRS for the full-season data (e.g. for rookie year samples), but MoV for the partial season data (e.g. for mid-season trades). If anyone has a way of easily calculating SRS for mid-season samples, let me know. I would love to change the MoV data to SRS data for accuracy, but I was unable of finding an easy way to do this.
-Rosters: Since this is multi-year data, there are likely other roster changes that bias some of these samples. I would love to include notes of the major contextual changes (e.g. if other all-star players went in or out of the lineup at the same time, if there were coaching changes or expansion, maybe even a list of every 20+/25+ mpg player who changed rosters between the ‘with’ sample and the ‘without’ sample. But this takes a lot of work and I haven’t gotten around to this yet.
-New Players: If there are any other players people would like to add, let me know. Preferably people could help with the calculations rather than just asking me to do all the work for a long list of players.
-Alternate Values: I added alternate values that seemed relevant, but the list is incomplete. There are likely other Alternate Values worth adding.
-Mistakes: I did this all by hand, so it’s possible there’s a typo or a small mistake. Don’t hesitate to (kindly) let me know and I can fix it.
If people want to help make these changes, don’t hesitate to reach out or post in the comments. Let me know if you have any other ideas that might help this database!

…

Bill Russell
-1956–57 Boston: 4.77 with, 1.68 without. Total change: +3.09 [Rookie year]
-1957–58 Boston: 6.05 with, 2.74 without. Total change: +3.31 [Injury year. *Note: 27 game off sample, included to give Russell a third sample]
-1969–70 Boston: 5.35 with, -1.59 without. Total change: +6.94 [Retirement]
Career Average: +4.45
10-year prime: +6.94 (1 sample 1960–69, +5.13 in 13 years 1957–1969)
Non-prime average: +3.20

Wilt Chamberlain
-1959–60 Warriors: 2.27 SRS with, -2.29 SRS without. Total change: +5.06 [Rookie year]
-1965 Warriors: -4.97 MoV with, -7.26 MoV without. Total change: +2.29. [trade, leaving Warriors]
-1965 76ers: 0.29 MoV with, -0.49 MoV without. Total change: +0.78. [trade, joining 76ers]
*Alternate Value: 1964–65 Warriors: +1.44 MoV with, -7.26 MoV without. Total change: +8.7 [Health Adjustment: Alternate value takes longer ‘with’ sample to correct for Wilt playing injured in 65.]
-1968–69 76ers: 7.96 SRS with, 4.79 SRS without. Total change: +3.17 [Traded, leaving 76ers]
*Alternate Value: 1965–66 76ers: +3.0 MoV with, -0.49 MoV without. Total change: +3.49 [Health Adjustment: Alternate value takes longer ‘with’ to correct for Wilt/teammates were playing injured]
-1968–69 Lakers: 3.84 SRS with, 4.99 SRS without. Total change: -1.15. [Traded, joining Lakers]
-1969-70 Lakers: 3.64 MoV with, 1.94 MoV without. Total change: +1.7 [Injury year]
*Alternate Value: 1970–1971 Lakers: Total change: +1.2 [Alternate Years: Alternate value uses alternate pair of years to get sufficient ‘with’ sample.]
-1973–74 Lakers: 8.16 SRS with, 0.85 SRS without. Total change: +7.31. [Retirement]
*Alternate Value: 1973–74 Lakers: 8.16 with, 1.29 without. Total Change: +6.87 [Teammate Adjustment: Alternate Value using games with 74 West for ‘without’ sample].
Career Average: +3.98 (using Alternate Values to correct for health in the 1965 years)
10-year prime: +3.85 (1960–69)
Non-prime average: +4.29

Oscar Robertson
-1960–61 Royals: -3.04 with, -5.92 without. Total change: +2.88 [Rookie year]
-1970–71 Royals: -2.55 with, -2.96 without. Total change: +0.41 [Traded, leaving Royals]
-1970–71 Bucks: 11.91 with, 4.25 without. Total change: +7.66 [Traded, joining Bucks]
-1974–75 Bucks: 7.61 with, 0.25 without. Total change: +7.36 [Retirement]
Career Average: +4.58
10-year prime: +4.04 (1962–71 2 samples)
Non-prime average: +5.12

Jerry West
-1960-61 Lakers: -0.11 with, -4.14 without. Total change: +4.03 [Rookie year]
-1968 Lakers: 8.86 with, 0.26 without. Total change: +8.6 [Injury year]
-1973–74 Lakers: 1.29 with, 0.61 without. Total change: +0.68 [Injury year]
-1974–75 Lakers: 1.29 with, -3.94 without. Total change: +5.23 [Retirement]
Career Average: +4.64
10-year prime: +6.32 (2 samples in 12 seasons 1961–1972, +8.6 in 1 sample 1963–1972)
Non-prime average: +2.96

Karem Abdul-Jabbar
-1969–70 Bucks: 4.25 with, -5.07 without. Total change: +9.32 [Rookie year]
*Adjusted Value: 1969–70 Bucks: 4.25 with, -2.62 without. Total change: +6.87 [Teammate Adjustment: Adjusted Value corrects for games with Flynn Robinson/Zaid Abdul-Aziz in 1969. Does not correct for addition of Bob Deandridge or expansion in 69–70]
-1975–76 Bucks: 0.25 with, -1.55 without. Total change: +1.8 [Traded, leaving Bucks]
-1975–76 Lakers: 0.18 with, -3.94 without. Total change: +4.12 [Traded, joining Lakers]
-1989–90 Lakers: 6.38 with, 6.74 without. Total change: -0.36 [Retirement]
Career Average: +3.12
10-year prime: +4.26 (1970–1979)
Non-prime average: -0.36 (1 sample in retirement. 3.22 in 2 samples including rookie year)

Larry Bird
-1979–80 Celtics: 7.37 with, -4.78 without. Total change: +12.15 [Rookie year]
-1988–89 Celtics: 5.61 with, 1.39 without. Total change: +4.22 [Injury year]
*Alternate Value: 1989–90 Celtics: Total change: +1.5 [Alternate Years: Alternate Value uses 1990 instead of 1988]
-1992 Celtics: 5.82 with, 0.95 without. Total change: +4.87 [Injury year]
-1992-93 Celtics: 5.82 with, 0.93 without. Total change: +4.89 [Retirement]
Career Average: +6.53
10-year prime: +8.19 (2 samples 1980–91 )
Non-prime average: +4.88

Magic Johnson
-1979–80 Lakers: 5.4 with, 2.95 without. Total change: +2.45 [Rookie year]
-1981 Lakers: 6.27 with, 1.84 without. Total change: +4.43 [Injury year]
-1991–92 Lakers: 6.73 with, -0.95 without. Total change: +7.68 [Injury year]
-1996 Lakers: 5.81 with, 3.58 without. Total change: +2.23 [Injury year]
-1996–97 Lakers: 5.81 with, 3.66 without. Total change: +2.15 [Retirement]
Career Average: +3.79
10-year prime: +6.06 (2 samples in 11 years 1981–1991, +7.68 in 10-year 1 sample)
Non-prime average: +2.28

Michael Jordan
-1984–85 Bulls: -0.5 with, -4.69 without. Total change: +4.19 [Rookie year]
-1986–87 Bulls: 0.38 with, -3.86 without. Total change: +4.24 [Injury year]
*Alternate Value: 1985–86 Bulls: Total change: +2.8 [Alternate Years: Alternate Value uses 1985 instead of 1987]
-1993–94 Bulls: 6.19 with, 2.87 without. Total change: +3.32 [Retirement]
*Alternate Value: 1992–95 Bulls: Total Change: +5.26 [Context Adjustment: Alternate Value using 1992–93 for the ‘with’ sample, since many have argued Bulls were coasting in 93. Note NBA expanded by 2 teams in 95]
-1995–96 Bulls: 10.96 with, 4.29 without. Total change: +6.67 [re-joining Bulls]
-1998–99 Bulls: 7.24 with, -8.58 without. Total change: +15.82 [Retirement]
*Alternate Value: 1998–99 Bulls: 8.55 with, -8.58 without. Total change: +11.28 [Teammate Adjustment: Alternate Value uses MoV with Pippen for ‘with’ sample, then subtract’s 98 Pippen’s 3.1 WOWY and 97 Rodman’s 2.75 WOWY].
-2001–02 Wizards: -1.57 with, -6.75 without. Total change: +5.18 [joining Wizards]
*Alternate Value: 2001–02 Wizards: Total change: +5.51 [Health Adjustment: Alternate Value only uses games Jordan played for ‘with’ sample]/
-2003–04 Wizards -1.47 with, -6.12 without. Total change: +4.65 [Retirement]
Career Average: +5.69 (using latter 2 alternate values)
10-year prime: +7.09 (1989–1998, +7.74 1989–1998 using alternate value for 1993 too)
Non-prime average: +4.65

Hakeem Olajuwon
-1984–85 Rockets: 1.38 with, -3.12 without. Total change: +4.5 [Rookie year]
-1991–92 Rockets: 1.7 with, -1.79 without. Total change: +3.49 [injury years]
-1998 Rockets: 0 with, -1.77 without. Total change: +1.77 [injury year]
-2000 Rockets: -3.7 with, 2.42 without. Total change: -6.12 [injury year]
-2001–02 Rockets: 2.71 with, -4.31 without. Total change: +7.02 [trade, leaving Rockets]
-2001–02 Raptors: -0.7 with, 1.69 without. Total change: -2.39 [trade, joining Raptors]
-2002–03 Raptors: -0.7 with, -6.1 without. Total change: +5.4 [Retirement]
Career Average: +1.95
10-year prime: +3.49 (1 sample 1986–95, +4.0 in 11 years 1985–95)
Non-prime average: +1.7

Shaquille O’Neal
-1992–93 Magic: 1.35 with, -6.52 without. Total change: +7.87 [Rookie year]
-1996–97 Magic: 5.4 with, -0.07 without. Total change: +5.47 [Traded, leaving Magic]
-1996–97 Lakers: 5.37 with, 3.73 without. Total change: +1.64 [Traded, joining Lakers]
*Adjusted Value: 1996–97 Lakers: Total Change: +2.22 [Teammate Adjustment: Alternate Value only uses games Magic did not play in 1996 for Shaq's 1996 'without' sample]
-2004–05 Lakers: 4.35 with, -2.32 without. Total change: +6.67 [Traded, leaving Lakers]
-2004–05 Heat: 5.76 with, -0.13 without. Total change: +5.89 [Traded, joining Heat]
-2007 Heat: 0.73 with, -2.48 without. Total change: +3.21 [Injury year]
-2008 Heat: 0.73 with, -8.53 without. Total change: +9.26 [Traded, leaving Heat]
-2008 Suns: 3.32 with, 5.94 without. Total change: -2.62 [Traded, joining Suns. *Note: 28 game ‘with’ sample, included for completeness.]
-2009–10 Suns: 3.32 with, 1.62 without. Total change: +1.7 [Traded, leaving Suns]
-2009–10 Cavs: 6.17 with, 8.68 without. Total change: -2.51 [Traded, joining Cavs]
-2010–11 Cavs: 6.17 with, -8.88 without. Total change: +15.05 [Traded, leaving Cavs]
*Adjusted Value: 2010–11 Cavs: Total change: +7.63 [Teammate Adjustment: Alternate Value subtracting 2011 Heat LeBron’s raw WOWY, using games with Varajao/Williams playing for ‘without’ sample]
-2010–11 Celtics: 4.83 with, 3.37 without. Total change: +1.46 [Traded, joining Celtics]
-2011–12 Celtics: 4.83 with, 2.26 without. Total change: +2.57 [Retirement]
Career Average: +3.74
10-year prime: +4.92 (1996–05, +4.59 1995–04)
Non-prime average: +3.17

Tim Duncan
-1997–98 Spurs: 3.3 with, -7.93 without. Total change: +11.23 [Rookie year]
*Alternate Value: 1997–98 Spurs: Total change: -0.31 [Teammate Adjustment: Alternate Value subtracts 96–97 Robinson’s 11.54 WOWY, which is calculated only when 96–97 Sean Elliot is playing]
-2004–05 Spurs: 8.89 with, 1.1 without. Total change: +7.79 [Injury year. *Only 29 ‘without’ games, included to give Duncan a third sample.]
-2016–17 Spurs: 10.28 with, 7.13 without. Total change: +3.15 [Retirement]
Career Average: +3.54 (with alternate value, which likely overcorrects)
10-year prime: +3.74 (1998–2007 in 2 samples with alternate value,
Non-prime average: +3.15 (1 sample)

Kevin Garnett
-1995–96 Timberwolves: -5.14 with, -8.22 without. Total change: +3.08 [Rookie year]
-2007–08 Timberwolves: -3.16 with, -6.26 without. Total change: +3.1 [Traded, leaving Timberwolves]
-2007–08 Celtics: 9.3 with, -3.7 without. Total change: +13.0 [Traded, joining Celtics]
*Alternate Value: 2007–08 Celtics: Total change: +9.30 [Teammate Adjustment: Alternate Value subtracting 07 Ray Allen’s 3.7 WOWY]
-2008–09 Celtics: 9.88 with, 5.33 without. Total change: +4.55 [Injury year]
*Alternate Value: 2009–10 Celtics: Total change: +3.98 [Alternate years: Alternate value using 2010 instead of 2008]
-2013–14 Celtics: -0.62 with, -4.97 without. Total change: +4.35 [Traded, leaving Celtics]
-2013–14 Nets: -1.57 with, 1.25 without. Total change: -2.82 [Traded, joining Nets]
-2015–16 Nets: -3.02 with, -2.73 without. Total change: -0.29 [Traded, leaving Nets]
-2015–16 Timberwolves: -2.21 with, -7.57 without. Total change: +5.36 [Traded, joining Timberwolves]
-2016 Timberwolves: -2.61 with, -4.34 without. Total change: +1.73 [Injury year]
-2016–17 Timberwolves: -2.61 with, -0.64 without. Total change: -1.97 [Retirement]
Career Average: +2.64 (with 2008 Alternate Value)
10-year prime: +5.65 (2000–09)
Non-prime average: +1.35

Kobe Bryant
-1996–97 Lakers: 3.66 with, 4.21 without. Total change: -0.55 [Rookie year]
-2000–01 Lakers: 6.51 with, 3.53 without. Total change: +2.98 [Injury year]
-2004–05 Lakers: 1.45 with, -3.42 without. Total change: +4.87 [Injury year]
-2013–14 Lakers: 0.81 with, -6.13 without. Total change: +6.94 [Injury year]
-2015 Lakers: -7 with, -6.72 without. Total change: -0.28 [Injury year]
-2016–17 Lakers: -6.29 with, -8.92 without. Total change: +2.63 [Retirement]
Career Average: +2.77
10-year prime: +3.93 (2 samples 2001–09)
Non-prime average: +2.19

LeBron James
-2003–04 Cavs: -3.07 with, -9.59 without. Total change: +6.52 [Rookie year]
-2010–11 Cavs: 6.17 with, -8.88 without. Total change: +15.05 [Traded, leaving Cavs]
*Adjusted Value: 2010–11 Cavs: Total Change: +10.94 [Teammate Adjustment: Alternate value subtracting 2011 Boston Shaq’s raw WOWY, using games with Varajao/Williams playing for ‘without’ sample]
-2010–11 Heat: 6.76 with, 1.99 without. Total change: +4.77 [Traded, joining Heat]
-2014–15 Heat: 4.15 with, -2.92 without. Total change: +7.07 [Traded, leaving Heat]
-2014–15 Cavs: 4.08 with, -3.86 without. Total change: +7.94 [Traded, joining Cavs]
-2018–19 Cavs: 0.59 with, -9.39 without. Total change: +9.98 [Traded, leaving Cavs]
-2018–19 Lakers: -1.33 with, -1.44 without. Total change: +0.11 [Traded, joining Lakers]
*Alternate Value: 2018–19 Lakers: Total change: +1.09 [Health Adjustment: Alternate value only uses when LeBron played for ‘with’ sample]
-2019–20 Lakers: 3.17 with, -3.78 without. Total change: +6.95 [Injury year]
-2021–22 Lakers: 1.25 with, -3.33 without. Total change: +4.58 [Injury year]
-2022–23 Lakers: -0.5 with, -2.8 without. Total change: +2.3 [Injury year]
Career Average: +6.21 (using alternate values)
10-year prime: +8.14 (2009–18)
Non-prime average: +4.29

Steph Curry
-2009–10 Warriors: -3.28 with, -3.8 without. Total change: +0.52 [Rookie year]
-2012 Warriors: -0.54 with, -5.28 without. Total change: +4.74 [Injury year]
-2018 Warriors: 9.71 with, -0.16 without. Total change: +9.87 [Injury year] ? (51 games)
-2020–21 Warriors: 0.4 with, -8.52 without. Total change: +8.92 [Injury year]
*Alternate Value: 2019–20 Warriors: Total change: +12.97 [Alternate years: Alternate Value using 2019 instead of 2021, subtracting 2021 Durant’s 1.97 WOWY. Klay’s 17–19 WOWY / 22 WOWY are both too noisy to use and at same time as other injuries.]
-2022–23 Warriors: 5.38 with, -1.0 without. Total: 6.38 [Injury year]
Career Average: +6.09
10-year prime: +8.39 (2014–2023)
Non-prime average: +2.63

Context behind various samples:

Spoiler:

…

It’s difficult to compare career averages in this data, given the noise of WOWY and the unequal sampling (e.g. some player’s samples are non-prime, others are near peak). Pitting players’ prime samples against each other is a bit fairer, although many players are still dealing with a small number of samples and the usual limitations of raw WOWY. You could also make different choices regarding whether to use each player’s Alternate Values or explore the necessary context behind each sample. But if only because it’s interesting, let’s see how players compare in their 10-year primes, in their non-prime years, and in their peak samples.
Note: For these numbers, I'm using the 'Alternate Value' whenever there's a health adjustment or teammate adjustment. When there's an 'Alternate Year' alternate value, I pick the set of years that's better (with the exception of Curry's 2019–20 alternate value, as this doesn't correct for Klay while 2020–21 does).

10-year primes:
-Steph Curry +8.39
-Larry Bird +8.19 (2 samples)
-LeBron James +8.14
-Michael Jordan +7.09 (1989–1998)
-Bill Russell +6.94 (1 sample)
-Jerry West +6.32 (2 samples in 12 years, +8.6 in 1 sample in 10 years)
-Magic Johnson +6.06 (2 samples in 11 years, +7.68 in 10-year 1 sample)
-Kevin Garnett +5.65
-Shaquille O’Neal + 5.06 (1996–05, +4.79 1995–04)
-Kareem Abdul-Jabbar +4.26
-Oscar Robertson +4.04 (2 samples)
-Hakeem Olajuwon +4.0 (2 samples in 11 years, +3.49 in 1 sample)
-Kobe Bryant +3.93 (2 samples)
-Wilt Chamberlain +3.85
-Tim Duncan +3.74 (2 samples)

Non Prime Years:
-Oscar Robertson +5.12 (2 samples)
-Larry Bird +4.88 (2 samples)
-Michael Jordan +4.65
-Wilt Chamberlain +4.29 (2 samples)
-LeBron James +4.29
-Kareem Abdul-Jabbar +3.22 (2 samples)
-Bill Russell +3.20 (2 samples)
-Shaquille O’Neal +3.17
-Tim Duncan +3.15 (1 sample)
-Jerry West +2.96 (2 samples)
-Steph Curry +2.63 (2 samples)
-Magic Johnson +2.28
-Kobe Bryant +2.19
-Hakeem Olajuwon +1.7
-Kevin Garnett +1.35

Peak years (samples over 8+ WOWY):
-1980 Bird: +12.15
-1998 Jordan: +11.28 (adjusted for Pippen/Rodman)
-2010 LeBron: +10.94 (adjusted for Shaq & injured Varejao/Williams)
-2018 LeBron: +9.98
-2018 Curry: +9.87
-2008 Garnett: +9.30 (adjusted for Allen)
-2008 Shaq: +9.26
-2021 Curry: +8.92
-1964 Wilt: +8.7 (adjusted for Wilt and his teammate’s health)
-1968 West: 8.6

...

Feel free to use this as a resource if you're comparing players using multi-year or large-sample WOWY. As I said above, this database may continue to expand or improve based on feedback, so don't hesitate to kindly reach out if you have any ideas. I look forward to reading your comments

lessthanjake · Post #2 » by **lessthanjake** » Mon Jul 24, 2023 11:19 pm

This is excellent—thank you for putting it together!

homecourtloss · Post #3 » by **homecourtloss** » Tue Jul 25, 2023 12:34 am

DraymondGold wrote:~The Multi-Year Large-Sample WOWY Database~

There has been much talk about WOWY in the recent 2023 Top 100 Project, so I’ve decided to start a database for multi-year, large sample raw WOWY. But first, a bit of background.

What is WOWY?
Spoiler:
WOWY, or “with or without you”, is a raw impact stat like on/off. But where on/off looks at the difference between possessions you’re on the court vs off the court within individual games, WOWY looks at games. It compares how a player’s team performs in games they play vs games they do not play. Traditionally this is measured using SRS (margin of victory, or MoV, adjusted for opponents) for accuracy, though you could also calculate a WOWY score using Margin of Victory.
Pros: 1) WOWY gets at wholistic value. It attempts to measure a player’s overall impact on a game. 2) Since it’s an impact stat, it may be better than box stats at measuring subtler non-box forms of impact. This might include rim deterrence and positioning on defense, off-ball motion on offense, or even on-court communication and leadership.  
Cons: 1) WOWY measures situational value. Just like raw on/off can be biased by lineups, raw WOWY can be biased by rosters. Certain rosters may be better equipped to replace a certain player’s role when they’re gone, while others may not have viable replacements. Certain coaches may be better at adjusting gameplans without a star player than others, which may inflate or deflate a WOWY rating. 2) WOWY is also context blind. Like raw on/off, raw WOWY does not adjust for teammates. If Player A is in for part of the season then out for the next, while their teammates are improving over time, this might dent Player A’s WOWY, because raw WOWY doesn’t know to adjust for the fact that Player A’s teammate are changing (or even that other players are coming in/out of the roster). Likewise, if a player is out for a few weeks then returns to games while still injured, this will lower their raw WOWY, because WOWY doesn’t know the player is injured. Likewise, 3) WOWY is super noisy. In other words, there’s a lot of uncertainty. Even full career averages can have large uncertainty ranges: even if Player A has +6 WOWY and Player B has +5 WOWY, they may not be far enough apart for us to say Player A is more valuable beyond our range of uncertainty. 4) Relatedly, WOWY often comes in much smaller sample sizes than on/off. In smaller samples, we expect this data to have greater uncertainty.

What types of WOWY data is there?
Spoiler:
The original WOWY metric was invented by Thinking Basketball. His database only included WOWY data from within single seasons. For example, it only included how team performance changed before and after mid-season trades, rather than trades during the off-season. This had the benefit of limiting how much team rosters and teammates changed when a player playing vs out (improving Con 2), at the cost of decreasing the sample size (worsening Con 4). He also applied a correction factor to correct for diminishing returns on good teams (“improving a +6 team to +10 is not typically the same ‘4 points’ as improving a -10 to -6.”). He then took a weighted average of this data to calculate a player’s career WOWY and prime WOWY.
Single-season WOWY database: https://thinkingbasketball.net/metrics/wowy-data/
Career and Prime WOWY database: http://www.backpicks.com/2016/08/24/i-historical-impact-wowy-score-update/
Prime WOWY scores for Top 40 Players: https://thinkingbasketball.net/2017/12/11/the-backpicks-goat-the-40-best-careers-in-nba-history/, available on each Top 40 profile.
You can also “adjust” raw WOWY, in the same way you can “adjust” raw plus minus to produce Adjusted Plus Minus (APM, or its cousin RAPM). This adjustment is traditionally called WOWYR. It has the added Pro of adjusting for teammate contributions (helping Con 2). There are alternative ways of calculating WOWYR (just like there are alternate ways of calculating RAPM), which produce slight differences. These alternates are called alt-WOWYR and alt-GPM. This data also only includes single-season WOWY as inputs. It does not look at changes in team performance before and after off-season trades, full-season injuries, retirements, etc.
Adjusted WOWY database: https://thinkingbasketball.net/metrics/wowyr/.

So what’s new with this WOWY data?
Where Thinking Basketball’s database only includes WOWY data from within a single season, here I’m calculating multi-year changes. This multi-year samples include changes to team performance over 1) rookie years, 2) mid-season/off-season trades, 3) long-term injuries, and 4) retirement years. This data has the benefit of being larger sample (helping Cons 3 and 4). However, since it’s over longer timescales, it’s more likely to be biased by other changes to the roster, improvement or drop-off from the player’s teammates, changes to the coaching staff or the league context, etc. (worsening Cons 1 and 2).
-Specific thresholds: I'm requiring at least 30 ‘with’ games and 30 ‘without’ games for a single team within a two-year span. It’s a somewhat arbitrary threshold, but the goal of this database is to have a sufficient sample size that there’s noise is minimized (as much as possible, although WOWY will certainly still have noise). I will use SRS for full-season data (e.g. for trades/injuries in the off season), MoV for mid-season data (e.g. for trades/injuries that occur mid-season). When relevant, I will include an ‘Alternate Value’ (e.g. if there are simple corrections for other roster changes, if you might use a different neighboring year, etc.).
-Sources: Basketball Reference for full-season SRS, Statmuse for mid-season MoV.
-The Players: Roughly the top 15 players. This isn’t meant to be a definitive list of the best WOWY scores or the true Top 15, just a list of the usual suspects that people are most frequently interested in. Those 15 players are… Bill Russell, Wilt Chamberlain, Oscar Robertson, Jerry West, Kareem Abdul-Jabbar, Larry Bird, Magic Johnson, Michael Jordan, Hakeem Olajuwon, Shaquille O’Neal, Tim Duncan, Kevin Garnett, Kobe Bryant, LeBron James, Steph Curry.

Future updates
This database is a work in progress, and may be expanded in the future. In particular:
-SRS: I used SRS for the full-season data (e.g. for rookie year samples), but MoV for the partial season data (e.g. for mid-season trades). If anyone has a way of easily calculating SRS for mid-season samples, let me know. I would love to change the MoV data to SRS data for accuracy, but I was unable of finding an easy way to do this.
-Rosters: Since this is multi-year data, there are likely other roster changes that bias some of these samples. I would love to include notes of the major contextual changes (e.g. if other all-star players went in or out of the lineup at the same time, if there were coaching changes or expansion, maybe even a list of every 20+/25+ mpg player who changed rosters between the ‘with’ sample and the ‘without’ sample. But this takes a lot of work and I haven’t gotten around to this yet.
-New Players: If there are any other players people would like to add, let me know. Preferably people could help with the calculations rather than just asking me to do all the work for a long list of players.
-Alternate Values: I added alternate values that seemed relevant, but the list is incomplete. There are likely other Alternate Values worth adding.
-Mistakes: I did this all by hand, so it’s possible there’s a typo or a small mistake. Don’t hesitate to (kindly) let me know and I can fix it.
If people want to help make these changes, don’t hesitate to reach out or post in the comments. Let me know if you have any other ideas that might help this database!

…

Bill Russell
-1956–57 Boston: 4.77 with, 1.68 without. Total change: +3.09 [Rookie year]
-1957–58 Boston: 6.05 with, 2.74 without. Total change: +3.31 [Injury year. *Note: 27 game off sample, included to give Russell a third sample]
-1969–70 Boston: 5.35 with, -1.59 without. Total change: +6.94 [Retirement]
Career Average: +4.45
10-year prime: +6.94 (1 sample 1960–69, +5.13 in 13 years 1957–1969)
Non-prime average: +3.20

Wilt Chamberlain
-1959–60 Warriors: 2.27 SRS with, -2.29 SRS without. Total change: +5.06 [Rookie year]
-1965 Warriors: -4.97 MoV with, -7.26 MoV without. Total change: +2.29. [trade, leaving Warriors]
-1965 76ers: 0.29 MoV with, -0.49 MoV without. Total change: +0.78. [trade, joining 76ers]
*Alternate Value: 1964–65 Warriors: +1.44 MoV with, -7.26 MoV without. Total change: +8.7 [Alternate value takes longer ‘with’ sample to correct for Wilt playing injured in 65.]
-1968–69 76ers: 7.96 SRS with, 4.79 SRS without. Total change: +3.17 [Traded, leaving 76ers]
*Alternate Value: 1965–66 76ers: +3.0 MoV with, -0.49 MoV without. Total change: +3.49 [Alternate value takes longer ‘with’ to correct for Wilt/teammates were playing injured]
-1968–69 Lakers: 3.84 SRS with, 4.99 SRS without. Total change: -1.15. [Traded, joining Lakers]
-1969-70 Lakers: 3.64 MoV with, 1.94 MoV without. Total change: +1.7 [Injury year]
*Alternate Value: 1970–1971 Lakers: Total change: +1.2 [Alternate value uses alternate pair of years to get sufficient ‘with’ sample.]
-1973–74 Lakers: 8.16 SRS with, 0.85 SRS without. Total change: +7.31. [Retirement]
*Alternate Value: 1973–74 Lakers: 8.16 with, 1.29 without. Total Change: +6.87 [Alternate Value using games with 74 West for ‘without’ sample].
Career Average: +3.98 (using Alternate Values to correct for health in the 1965 years)
10-year prime: +3.85 (1960–69)
Non-prime average: +4.29

Oscar Robertson
-1960–61 Royals: -3.04 with, -5.92 without. Total change: +2.88 [Rookie year]
-1970–71 Royals: -2.55 with, -2.96 without. Total change: +0.41 [Traded, leaving Royals]
-1970–71 Bucks: 11.91 with, 4.25 without. Total change: +7.66 [Traded, joining Bucks]
-1974–75 Bucks: 7.61 with, 0.25 without. Total change: +7.36 [Retirement]
Career Average: +4.58
10-year prime: +4.04 (1962–71 2 samples)
Non-prime average: +5.12

Jerry West
-1960-61 Lakers: -0.11 with, -4.14 without. Total change: +4.03 [Rookie year]
-1968 Lakers: 8.86 with, 0.26 without. Total change: +8.6 [Injury year]
-1973–74 Lakers: 1.29 with, 0.61 without. Total change: +0.68 [Injury year]
-1974–75 Lakers: 1.29 with, -3.94 without. Total change: +5.23 [Retirement]
Career Average: +4.64
10-year prime: +6.32 (2 samples in 12 seasons 1961–1972, +8.6 in 1 sample 1963–1972)
Non-prime average: +2.96

Karem Abdul-Jabbar
-1969–70 Bucks: 4.25 with, -5.07 without. Total change: +9.32 [Rookie year]
*Adjusted Value: 1969–70 Bucks: 4.25 with, -2.62 without. Total change: +6.87 [Adjusted Value corrects for games with Flynn Robinson/Zaid Abdul-Aziz in 1969. Does not correct for addition of Bob Deandridge or expansion in 69–70]
-1975–76 Bucks: 0.25 with, -1.55 without. Total change: +1.8 [Traded, leaving Bucks]
-1975–76 Lakers: 0.18 with, -3.94 without. Total change: +4.12 [Traded, joining Lakers]
-1989–90 Lakers: 6.38 with, 6.74 without. Total change: -0.36 [Retirement]
Career Average: +3.12
10-year prime: +4.26 (1970–1979)
Non-prime average: -0.36 (1 sample in retirement. 3.22 in 2 samples including rookie year)

Larry Bird
-1979–80 Celtics: 7.37 with, -4.78 without. Total change: +12.15 [Rookie year]
-1988–89 Celtics: 5.61 with, 1.39 without. Total change: +4.22 [Injury year]
*Alternate Value: 1989–90 Celtics: Total change: +1.5 [Alternate Value uses 1990 instead of 1988]
-1992 Celtics: 5.82 with, 0.95 without. Total change: +4.87 [Injury year]
-1992-93 Celtics: 5.82 with, 0.93 without. Total change: +4.89 [Retirement]
Career Average: +6.53
10-year prime: +8.19 (2 samples 1980–91 )
Non-prime average: +4.88

Magic Johnson
-1979–80 Lakers: 5.4 with, 2.95 without. Total change: +2.45 [Rookie year]
-1981 Lakers: 6.27 with, 1.84 without. Total change: +4.43 [Injury year]
-1991–92 Lakers: 6.73 with, -0.95 without. Total change: +7.68 [Injury year]
-1996 Lakers: 5.81 with, 3.58 without. Total change: +2.23 [Injury year]
-1996–97 Lakers: 5.81 with, 3.66 without. Total change: +2.15 [Retirement]
Career Average: +3.79
10-year prime: +6.06 (2 samples in 11 years 1981–1991, +7.68 in 10-year 1 sample)
Non-prime average: +2.28

Michael Jordan
-1984–85 Bulls: -0.5 with, -4.69 without. Total change: +4.19 [Rookie year]
-1986–87 Bulls: 0.38 with, -3.86 without. Total change: +4.24 [Injury year]
*Alternate Value: 1985–86 Bulls: Total change: +2.8 [Alternate Value uses 1985 instead of 1987]
-1993–94 Bulls: 6.19 with, 2.87 without. Total change: +3.32 [Retirement]
*Alternate Value: 1992–95 Bulls: Total Change: +5.26 [Alternate Value using 1992–93 for the ‘with’ sample, since many have argued Bulls were coasting in 93]
-1995–96 Bulls: 10.96 with, 4.29 without. Total change: +6.67 [re-joining Bulls]
-1998–99 Bulls: 7.24 with, -8.58 without. Total change: +15.82 [Retirement]
*Alternate Value: 1998–99 Bulls: 8.55 with, -8.58 without. Total change: +11.28 [Alternate Value uses MoV with Pippen for ‘with’ sample, then subtract’s 98 Pippen’s 3.1 WOWY and 97 Rodman’s 2.75 WOWY].
-2001–02 Wizards: -1.57 with, -6.75 without. Total change: +5.18 [joining Wizards]
*Alternate Value: 2001–02 Wizards: Total change: +5.51 [Alternate Value only uses games Jordan played for ‘with’ sample]/
-2003–04 Wizards -1.47 with, -6.12 without. Total change: +4.65 [Retirement]
Career Average: +5.69 (using latter 2 alternate values)
10-year prime: +7.09 (1989–1998, +7.74 1989–1998 using alternate value for 1993 too)
Non-prime average: +4.65

Hakeem Olajuwon
-1984–85 Rockets: 1.38 with, -3.12 without. Total change: +4.5 [Rookie year]
-1991–92 Rockets: 1.7 with, -1.79 without. Total change: +3.49 [injury years]
-1998 Rockets: 0 with, -1.77 without. Total change: +1.77 [injury year]
-2000 Rockets: -3.7 with, 2.42 without. Total change: -6.12 [injury year]
-2001–02 Rockets: 2.71 with, -4.31 without. Total change: +7.02 [trade, leaving Rockets]
-2001–02 Raptors: -0.7 with, 1.69 without. Total change: -2.39 [trade, joining Raptors]
-2002–03 Raptors: -0.7 with, -6.1 without. Total change: +5.4 [Retirement]
Career Average: +1.95
10-year prime: +3.49 (1 sample 1986–95, +4.0 in 11 years 1985–95)
Non-prime average: +1.7

Shaquille O’Neal
-1992–93 Magic: 1.35 with, -6.52 without. Total change: +7.87 [Rookie year]
-1996–97 Magic: 5.4 with, -0.07 without. Total change: +5.47 [Traded, leaving Magic]
-1996–97Lakers: 3.66 with, 4.21 without. Total change: -0.55 [Traded, joining Lakers]
-2004–05 Lakers: 4.35 with, -2.32 without. Total change: +6.67 [Traded, leaving Lakers]
-2004–05Heat: 5.76 with, -0.13 without. Total change: +5.89 [Traded, joining Heat]
-2007 Heat: 0.73 with, -2.48 without. Total change: +3.21 [Injury year]
-2008 Heat: 0.73 with, -8.53 without. Total change: +9.26 [Traded, leaving Heat]
-2008 Suns: 3.32 with, 5.94 without. Total change: -2.62 [Traded, joining Suns. *Note: 28 game ‘with’ sample, included for completeness.]
-2009–10 Suns: 3.32 with, 1.62 without. Total change: +1.7 [Traded, leaving Suns]
-2009–10 Cavs: 6.17 with, 8.68 without. Total change: -2.51 [Traded, joining Cavs]
-2010–11 Cavs: 6.17 with, -8.88 without. Total change: +15.05 [Traded, leaving Cavs]
*Adjusted Value: 2010–11 Cavs: Total change: +7.63 [Alternate Value subtracting 2011 Heat LeBron’s raw WOWY, using games with Varajao/Williams playing for ‘without’ sample]
-2010–11 Celtics: 4.83 with, 3.37 without. Total change: +1.46 [Traded, joining Celtics]
-2011–12 Celtics: 4.83 with, 2.26 without. Total change: +2.57 [Retirement]
Career Average: +3.54
10-year prime: +4.37 (1996–95, +3.86 1995–04)
Non-prime average: +3.17

Tim Duncan
-1997–98 Spurs: 3.3 with, -7.93 without. Total change: +11.23 [Rookie year]
*Alternate Value: 1997–98 Spurs: Total change: -0.31 [Alternate Value subtracts 96–97 Robinson’s 11.54 WOWY, which is calculated only when 96–97 Sean Elliot is playing]
-2004–05 Spurs: 8.89 with, 1.1 without. Total change: +7.79 [Injury year. *Only 29 ‘without’ games, included to give Duncan a third sample.]
-2016–17 Spurs: 10.28 with, 7.13 without. Total change: +3.15 [Retirement]
Career Average: +3.54 (with alternate value, which likely overcorrects)
10-year prime: +3.74 (1998–2007 in 2 samples with alternate value,
Non-prime average: +3.15 (1 sample)

Kevin Garnett
-1995–96 Timberwolves: -5.14 with, -8.22 without. Total change: +3.08 [Rookie year]
-2007–08 Timberwolves: -3.16 with, -6.26 without. Total change: +3.1 [Traded, leaving Timberwolves]
-2007–08 Celtics: 9.3 with, -3.7 without. Total change: +13.0 [Traded, joining Celtics]
*Alternate Value: 2007–08 Celtics: Total change: +9.30 [Alternate Value subtracting 07 Ray Allen’s 3.7 WOWY]
-2008–09 Celtics: 9.88 with, 5.33 without. Total change: +4.55 [Injury year]
*Alternate Value: 2009–10 Celtics: Total change: +3.98 [Alternate value using 2010 instead of 2008]
-2013–14 Celtics: -0.62 with, -4.97 without. Total change: +4.35 [Traded, leaving Celtics]
-2013–14 Nets: -1.57 with, 1.25 without. Total change: -2.82 [Traded, joining Nets]
-2015–16 Nets: -3.02 with, -2.73 without. Total change: -0.29 [Traded, leaving Nets]
-2015–16 Timberwolves: -2.21 with, -7.57 without. Total change: +5.36 [Traded, joining Timberwolves]
-2016 Timberwolves: -2.61 with, -4.34 without. Total change: +1.73 [Injury year]
-2016–17 Timberwolves: -2.61 with, -0.64 without. Total change: -1.97 [Retirement]
Career Average: +2.64 (with 2008 Alternate Value)
10-year prime: +5.65 (2000–09)
Non-prime average: +1.35

Kobe Bryant
-1996–97 Lakers: 3.66 with, 4.21 without. Total change: -0.55 [Rookie year]
-2000–01 Lakers: 6.51 with, 3.53 without. Total change: +2.98 [Injury year]
-2004–05 Lakers: 1.45 with, -3.42 without. Total change: +4.87 [Injury year]
-2013–14 Lakers: 0.81 with, -6.13 without. Total change: +6.94 [Injury year]
-2015 Lakers: -7 with, -6.72 without. Total change: -0.28 [Injury year]
-2016–17 Lakers: -6.29 with, -8.92 without. Total change: +2.63 [Retirement]
Career Average: +2.77
10-year prime: +3.93 (2 samples 2001–09)
Non-prime average: +2.19

LeBron James
-2003–04 Cavs: -3.07 with, -9.59 without. Total change: +6.52 [Rookie year]
-2010–11 Cavs: 6.17 with, -8.88 without. Total change: +15.05 [Traded, leaving Cavs]
*Adjusted Value: 2010–11 Cavs: Total Change: +10.94 [Alternate value subtracting 2011 Boston Shaq’s raw WOWY, using games with Varajao/Williams playing for ‘without’ sample]
-2010–11 Heat: 6.76 with, 1.99 without. Total change: +4.77 [Traded, joining Heat]
-2014–15 Heat: 4.15 with, -2.92 without. Total change: +7.07 [Traded, leaving Heat]
-2014–15 Cavs: 4.08 with, -3.86 without. Total change: +7.94 [Traded, joining Cavs]
-2018–19 Cavs: 0.59 with, -9.39 without. Total change: +9.98 [Traded, leaving Cavs]
-2018–19 Lakers: -1.33 with, -1.44 without. Total change: +0.11 [Traded, joining Lakers]
*Alternate Value: 2018–19 Lakers: Total change: +1.09 [Alternate value only uses when LeBron played for ‘with’ sample]
-2019–20 Lakers: 3.17 with, -3.78 without. Total change: +6.95 [Injury year]
-2021–22 Lakers: 1.25 with, -3.33 without. Total change: +4.58 [Injury year]
-2022–23 Lakers: -0.5 with, -2.8 without. Total change: +2.3 [Injury year]
Career Average: +6.21 (using alternate values)
10-year prime: +8.14 (2009–18)
Non-prime average: +4.29

Steph Curry
-2009–10 Warriors: -3.28 with, -3.8 without. Total change: +0.52 [Rookie year]
-2012 Warriors: -0.54 with, -5.28 without. Total change: +4.74 [Injury year]
-2018 Warriors: 9.71 with, -0.16 without. Total change: +9.87 [Injury year] ? (51 games)
-2020–21 Warriors: 0.4 with, -8.52 without. Total change: +8.92 [Injury year]
*Alternate Value: 2019–20 Warriors: Total change: +12.97 [Alternate Value using 2019 instead of 2021, subtracting 2021 Durant’s 1.97 WOWY. Klay’s 17–19 WOWY / 22 WOWY are both too noisy to use and at same time as other injuries.]
-2022–23 Warriors: 5.38 with, -1.0 without. Total: 6.38 [Injury year]
Career Average: +6.09
10-year prime: +8.39 (2014–2023)
Non-prime average: +2.63

Context behind various samples:
Spoiler:
Here will be a growing list of deep dives into the specific context behind certain raw WOWY samples. For example, they might provide context behind some of the ‘Adjusted Values’. Take Wilt’s un-adjusted WOWY value in 1965. It turns out Wilt and his teammates were injured at the time, likely limiting Wilt’s raw WOWY (see discussion below). This motivates the Adjusted Value you’ll find above. If people find other deep dives that are worth adding to this growing list, let me know!

1964–1966 Wilt: https://forums.realgm.com/boards/viewtopic.php?p=107709117#p107709117.
-1965 Wilt had either a heart attack or pancreatitis that some Doctors thought should keep him out for the season. He played through injury this year
-1965 Wilt’s new 76ers teammates Greer, Jackson, and Costello got injured not long after Wilt was traded to the 76ers.
-1965 Warriors had Nate Thurmond as Wilt’s replacement center.
This likely explains why Wilt's WOWY is so low in the middle of his prime. The Alternate Value takes a longer sample to try to correct for this.

1970 Kareem’s Bucks: https://forums.realgm.com/boards/viewtopic.php?p=104187436#p104187436, bottom of the post.
-WOWY underrates 1969 Bucks: Gained Flynn Robinson (3rd best player) and rookie Zaid Abdul-Aziz (5th in minutes) part way through season. Greg Smith (one of their best players) was a rookie
-WOWY overrates 1970 Bucks’ jump with Kareem: Bob Deandridge (3rd best player in 71) joined along with Kareem. Rookies Zaid Abdul-Aziz and Greg Smith likely improved too.
-Team context: Expansion tends to produce extra bad teams in their first years of existence, and the best teams tend to have overrated SRS because they feast on the bad teams. 1969 Bucks was the first year of existence (totally new team, first-time coach who was a player immediately prior). 1970 Bucks had higher SRS in expansion year.

1993 Jordan’s Bulls: https://forums.realgm.com/boards/viewtopic.php?p=107591289#p107591289 Deep dive into the contextual changes from 1993 to 1994, with debate in the following posts.
-Raw WOWY underrates 1993 with jordan: Pippen and Grant both had down years in efficiency and in turnovers in 1993, compared to any other years (e.g. bounce back year in 1994). This supports the idea that the 93 Bulls were coasting or at least their performance wasn’t representative of Pippen/Grant’s true long-term value.
-Raw WOWY underrates the drop in 1994: The Bulls improved most of their bench. Toni Kukoc replaced a worse player, Steve Kerr replaced worse players, Bill Wennington replaced worse player.
-Playoff WOWY overrates the Bulls in the 1994 Playoffs, as their first round series had opponents Nance, Daugherty, and HR Williams all out for the playoffs.
-Comparing 1994 to 1992, Jordan would have a +7.2 SRS change, +5.0 SRS change accounting for healthy 1992 Bulls, assuming none of the above roster changes are limiting Jordan’s WOWY, and not curving raw WOWY up to account for diminishing returns on a good team.

2010-11 LeBron’s Cavaliers: http://20secondtimeout.blogspot.com/2011/07/analyzing-collapse-of-2010-11-cleveland.html?m=1 / [url]ESPNwww.espn.comCleveland Cavaliers: Why the Cavs have imploded in LeBron James' wake[/url]. Some
-Raw WOWY exaggerates the 2011 cavs drop because of LeBron, as it doesn’t include:
-1) They also traded: Zydrunas Ilgauskas, Shaquille O'Neal and Delonte West
-2) They lost their coach: COTY Mike Brown, and GM Danny Ferry
-3) They had injuries: “Antawn Jamison and Mo Williams began the season with nagging injuries but then the death blow to any hope for the Cavs arrived when Anderson Varejao--the team's only credible inside player--suffered a season-ending injury.
…

It’s difficult to compare career averages in this data, given the noise of WOWY and the unequal sampling (e.g. some player’s samples are non-prime, others are near peak). Pitting players’ prime samples against each other is a bit fairer, although many players are still dealing with a small number of samples and the usual limitations of raw WOWY. You could also make different choices regarding whether to use each player’s Alternate Values or explore the necessary context behind each sample. But if only because it’s interesting, let’s see how players compare in their 10-year primes, in their non-prime years, and in their peak samples (using my choices for the most reasonable pick between player's original sample and 'alternate value' sample).

10-year primes:
-Steph Curry +8.39
-Larry Bird +8.19 (2 samples)
-LeBron James +8.14
-Michael Jordan +7.09 (1989–1998)
-Bill Russell +6.94 (1 sample)
-Jerry West +6.32 (2 samples in 12 years, +8.6 in 1 sample in 10 years)
-Magic Johnson +6.06 (2 samples in 11 years, +7.68 in 10-year 1 sample)
-Kevin Garnett +5.65
-Shaquille O’Neal +4.37 (1996–95, +3.86 1995–04)
-Kareem Abdul-Jabbar +4.26
-Oscar Robertson +4.04 (2 samples)
-Hakeem Olajuwon +4.0 (2 samples in 11 years, +3.49 in 1 sample)
-Kobe Bryant +3.93 (2 samples)
-Wilt Chamberlain +3.85
-Tim Duncan +3.74 (2 samples)

Non Prime Years:
-Oscar Robertson +5.12 (2 samples)
-Larry Bird +4.88 (2 samples)
-Michael Jordan +4.65
-Wilt Chamberlain +4.29 (2 samples)
-LeBron James +4.29
-Kareem Abdul-Jabbar +3.22 (2 samples)
-Bill Russell +3.20 (2 samples)
-Shaquille O’Neal +3.17
-Tim Duncan +3.15 (1 sample)
-Jerry West +2.96 (2 samples)
-Steph Curry +2.63 (2 samples)
-Magic Johnson +2.28
-Kobe Bryant +2.19
-Hakeem Olajuwon +1.7
-Kevin Garnett +1.35

Peak years (samples over 8+ WOWY):
-1980 Bird: +12.15
-1998 Jordan: +11.28 (adjusted for Pippen/Rodman)
-2010 LeBron: +10.94 (adjusted for Shaq & injured Varejao/Williams)
-2018 LeBron: +9.98
-2018 Curry: +9.87
-2008 Garnett: +9.30 (adjusted for Allen)
-2008 Shaq: +9.26
-2021 Curry: +8.92
-1964 Wilt: +8.7 (adjusted for Wilt and his teammate’s health)
-1968 West: 8.6

...

Feel free to use this as a resource if you're comparing players using multi-year or large-sample WOWY. As I said above, this database may continue to expand or improve based on feedback, so don't hesitate to kindly reach out if you have any ideas. I look forward to reading your comments

What are the numbers without any “adjusted” or “alternate” values?

OhayoKD · Post #4 » by **OhayoKD** » Tue Jul 25, 2023 1:58 am

Appreicate the effort here. A few notes.

DraymondGold wrote:~The Multi-Year Large-Sample WOWY Database~
elatedly, WOWY often comes in much smaller sample sizes than on/off. In smaller samples, we expect this data to have greater uncertainty.

Depends on what you're using. If you are using "indirect" you get the largest possible samples in terms of off. Rest seems fine. Various adjustments like "full-strenght" and the like can be used to reduce noise.

For those with a "championship" focus, general league-wide context matters too. +4 in a league or era of +2's > +6 in a league or era of +8's(important with Russell, Wilt and Kareem specifically).

Lower and upper-bounds are a useful concept too I think. Whatever you think of 94, 1995 drop is probably underrating Jordan's cast, while the 71 Celtics drop relative to 69 probably overrates Russell's.

2010-11 LeBron’s Cavaliers: http://20secondtimeout.blogspot.com/2011/07/analyzing-collapse-of-2010-11-cleveland.html?m=1 / [url]ESPNwww.espn.comCleveland Cavaliers: Why the Cavs have imploded in LeBron James' wake[/url]. Some
-Raw WOWY exaggerates the 2011 cavs drop because of LeBron, as it doesn’t include:

We have a 21-game sample from 2011 where the Cavs played the same starters Lebron was playing with. They posted a 18-win srs. Adjusted value being "+7" seems off. I would just use that 21-game sample.

Can do a lower-bound(likely undersells Lebron) adjustment with 2004 Lebron where you take boozer's net-rating from the next year and subtract it from the jump the 2004 cavs see with Lebron. Boozer was actually on the 2003 roster(and probably had improved by 2005) but iirc that lowers the jump to +4. Can check later

1970 Kareem’s Bucks: https://forums.realgm.com/boards/viewtopic.php?p=104187436#p104187436, bottom of the post.
-WOWY underrates 1969 Bucks: Gained Flynn Robinson (3rd best player) and rookie Zaid Abdul-Aziz (5th in minutes) part way through season. Greg Smith (one of their best players) was a rookie
-WOWY overrates 1970 Bucks’ jump with Kareem: Bob Deandridge (3rd best player in 71) joined along with Kareem. Rookies Zaid Abdul-Aziz and Greg Smith likely improved too.
-Team context: Expansion tends to produce extra bad teams in their first years of existence, and the best teams tend to have overrated SRS because they feast on the bad teams. 1969 Bucks was the first year of existence (totally new team, first-time coach who was a player immediately prior). 1970 Bucks had higher SRS in expansion year.

Reverse applies to Kareem's Lakers. 75 roster loses players and SRS is generally lower as of 77 and beyond.

*Alternate Value: 1992–95 Bulls: Total Change: +5.26 [Alternate Value using 1992–93 for the ‘with’ sample, since many have argued Bulls were coasting in 93]
-1995–96 Bulls: 10.96 with, 4.29 without. Total change: +6.67 [re-joining Bulls]

95-96 doesn't really fit the rest of what is included here. Jordan was on the team. If you're looking fora 96 extrap, using 94 to 96 makes sense. If "player improves" is being included, then 70 -> 71 kareem, 08 -> 09 Lebron, 2014 -> 2015 steph all seem equally valid as inclusions

I'd also add Jordan is replaced with pete myers which is a pretty big downgrade but I tend to just use 92 anyway. Am curious how exactly you're adjusting for 1999.

migya · Post #5 » by **migya** » Tue Jul 25, 2023 2:19 am

Do David Robinson

DraymondGold · Post #6 » by **DraymondGold** » Tue Jul 25, 2023 2:32 am

homecourtloss wrote:
DraymondGold wrote:[...

What are the numbers without any “adjusted” or “alternate” values?

Literally just raw WOWY, that's it. You can find an expanded definition with links to original sources under the "What is WOWY" spoiler in the OP.

But to be more specific: raw WOWY is just the change in team performance when a player plays in a game vs when they miss a game over a period of time. So for example, the formula for Raw WOWY is: Raw WOWY = [SRS (or Margin of Victory per game) when a player plays games for a certain team] - [SRS (or Margin of Victory per game) when a player misses games for a certain team].

Let's look at Michael Jordan, just to give a few examples.

Example 1: Michael Jordan Rookie Season sample. These are full year samples. Jordan played 0 games in 1984, and played 82 games in 1985. So we can use SRS to evaluate the change due to adding Jordan (and due to other roster changes, players aging, general variance, etc.).

So 1984–85 Jordan's raw WOWY = [SRS in season when he played] - [SRS in season when he didn't] = [1985 Bulls' SRS of -0.50] – [1984 Bulls' SRS of -4.69 = +4.19

Example 2: Michael Jordan's 1986 Injury. right off the bat, this sample is different. Jordan didn’t miss the full year, he missed only part of the year (missed 64 games, played 18). Like I said in OP, I’m using Margin of Victory per game for sub-season samples. So we get MoV for 1986 Bulls in games with Jordan and without Jordan.

But we reach a second problem: I set a minimum threshold of 30 games with and without, while we only have 18 games with. So we need to go to a neighboring year. Since Jordan was playing at the end of the 1986 season, then 1987 is closer for a “with” sample, so let’s go with 1987 (which has the added benefit of being a non-rookie year, which seems fairer to Jordan). We get the SRS or MoV for 1987 Bulls in games with and without Jordan (in this case, he played the full season so we get the team SRS… if he missed a non-negligible number of games we’d get the “with Jordan” MoV and “without Jordan” MoV).

Then we calculate the WOWY: 1986-87 WOWY = [average 86-87 Bulls MoV or SRS in games with Jordan playing] - [average 86-87 Bulls MoV or SRS in games without Jordan playing] = [0.38 with] - [-3.86 without] = 4.24.

But I made a somewhat arbitrary choice here. I chose to use 1987 for the second sample, rather than the other neighboring year (1985). I thought it was a better choice, but it was still arbitrary. So we can do the same process using 1985 instead of 1986 to get the “Alternate Value”. I include the alternate value for completeness.

Example 3: Michael Jordan's 1998 Retirement.
On the face of it, this is like Example 1. Jordan played the entire 1998 year and missed the entire 1999 year, so we can sue SRS for both. And that’s what I do for the standard value.

So why do we have an “Alternate Value” here? Well, there were pretty major other roster changes that we might correct for. Pippen and Rodman also left the Bulls at the same time Jordan retired. So this inspires the Alternate Value.

I check if there’s a raw WOWY sample for these two players within a year or two. I make sure it has a non-negligible sample size and doesn’t look like noise (e.g. it doesn’t put Pippen’s value at -10.0 which would obviously wrong). And as it turns out, there is!

Using the same process, we can get a raw WOWY for 1998 Pippen. We look at the 1998 Bulls MoV with and without Pippen, and find Pippen has a raw WOWY of +3.1. Rodman didn’t miss enough games in 98, but he did miss more in 97, and we find he has a raw WOWY of +2.75. So to isolate Jordan’s (approximate) WOWY, we take the full WOWY change in the 98 Bulls’ MoV with Pippen playing compared to the 99 Bulls’ SRS without any of them, then subtract Pippen and Rodman’s raw WOWY. The remainder is Jordan’s approximate value (or other contextual changes from the roster, coaching staff, etc).

So that’s the alternate value in 1998

…

If you have any other questions, feel free to check the article I linked in the “What is WOWY” spoiler above, or just ask if there’s any samples you’re curious about!

homecourtloss · Post #7 » by **homecourtloss** » Tue Jul 25, 2023 2:48 am

DraymondGold wrote:
homecourtloss wrote:
DraymondGold wrote:[...

What are the numbers without any “adjusted” or “alternate” values?
Literally just raw WOWY, that's it. You can find an expanded definition with links to original sources under the "What is WOWY" spoiler in the OP.

But to be more specific: raw WOWY is just the change in team performance when a player plays in a game vs when they miss a game over a period of time. So for example, the formula for Raw WOWY is: Raw WOWY = [SRS (or Margin of Victory per game) when a player plays games for a certain team] - [SRS (or Margin of Victory per game) when a player misses games for a certain team].

Let's look at Michael Jordan, just to give a few examples.

Example 1: Michael Jordan Rookie Season sample. These are full year samples. Jordan played 0 games in 1984, and played 82 games in 1985. So we can use SRS to evaluate the change due to adding Jordan (and due to other roster changes, players aging, general variance, etc.).

So 1984–85 Jordan's raw WOWY = [SRS in season when he played] - [SRS in season when he didn't] = [1985 Bulls' SRS of -0.50] – [1984 Bulls' SRS of -4.69 = +4.19

Example 2: Michael Jordan's 1986 Injury. right off the bat, this sample is different. Jordan didn’t miss the full year, he missed only part of the year (missed 64 games, played 18). Like I said in OP, I’m using Margin of Victory per game for sub-season samples. So we get MoV for 1986 Bulls in games with Jordan and without Jordan.

But we reach a second problem: I set a minimum threshold of 30 games with and without, while we only have 18 games with. So we need to go to a neighboring year. Since Jordan was playing at the end of the 1986 season, then 1987 is closer for a “with” sample, so let’s go with 1987 (which has the added benefit of being a non-rookie year, which seems fairer to Jordan). We get the SRS or MoV for 1987 Bulls in games with and without Jordan (in this case, he played the full season so we get the team SRS… if he missed a non-negligible number of games we’d get the “with Jordan” MoV and “without Jordan” MoV).

Then we calculate the WOWY: 1986-87 WOWY = [average 86-87 Bulls MoV or SRS in games with Jordan playing] - [average 86-87 Bulls MoV or SRS in games without Jordan playing] = [0.38 with] - [-3.86 without] = 4.24.

But I made a somewhat arbitrary choice here. I chose to use 1987 for the second sample, rather than the other neighboring year (1985). I thought it was a better choice, but it was still arbitrary. So we can do the same process using 1985 instead of 1986 to get the “Alternate Value”. I include the alternate value for completeness.

Example 3: Michael Jordan's 1998 Retirement.
On the face of it, this is like Example 1. Jordan played the entire 1998 year and missed the entire 1999 year, so we can sue SRS for both. And that’s what I do for the standard value.

So why do we have an “Alternate Value” here? Well, there were pretty major other roster changes that we might correct for. Pippen and Rodman also left the Bulls at the same time Jordan retired. So this inspires the Alternate Value.

I check if there’s a raw WOWY sample for these two players within a year or two. I make sure it has a non-negligible sample size and doesn’t look like noise (e.g. it doesn’t put Pippen’s value at -10.0 which would obviously wrong). And as it turns out, there is!

Using the same process, we can get a raw WOWY for 1998 Pippen. We look at the 1998 Bulls MoV with and without Pippen, and find Pippen has a raw WOWY of +3.1. Rodman didn’t miss enough games in 98, but he did miss more in 97, and we find he has a raw WOWY of +2.75. So to isolate Jordan’s (approximate) WOWY, we take the full WOWY change in the 98 Bulls’ MoV with Pippen playing compared to the 99 Bulls’ SRS without any of them, then subtract Pippen and Rodman’s raw WOWY. The remainder is Jordan’s approximate value (or other contextual changes from the roster, coaching staff, etc).

So that’s the alternate value in 1998

…

If you have any other questions, feel free to check the article I linked in the “What is WOWY” spoiler above, or just ask if there’s any samples you’re curious about!

I am quite familiar with WOWY. Perhaps I wasn’t clear—I was asking for the career, 10 year, peak, etc., numbers that you listed for the players without the “alternate” and “adjusted” numbers.

DraymondGold · Post #8 » by **DraymondGold** » Tue Jul 25, 2023 3:16 am

OhayoKD wrote:Appreicate the effort here. A few notes.
DraymondGold wrote:~The Multi-Year Large-Sample WOWY Database~
elatedly, WOWY often comes in much smaller sample sizes than on/off. In smaller samples, we expect this data to have greater uncertainty.

Depends on what you're using. If you are using "indirect" you get the largest possible samples in terms of off. Rest seems fine. Various adjustments like "full-strenght" and the like can be used to reduce noise.

The indirect stuff definitely helps give a larger off value, and 'full strength' adjustments definitely help reduce noise (though not entirely). So agreed there

On/off is still a larger sample regardless. To be more explicit, on/off depends on possession samples and WOWY depends on game samples, and there are *far* more 'on' possessions and 'off' possessions than there are 'with a player' games and 'without a player' games. Stars pretty easily play over a thousand possessions in a season, and they're of course don't play a thousand games in one or two seasons (that would be a crazy season! :lol:

)

For those with a "championship" focus, general league-wide context matters too. +4 in a league or era of +2's > +6 in a league or era of +8's(important with Russell, Wilt and Kareem specifically).

Sure. Basically this is saying there's raw WOWY and then raw WOWY standard deviations, right?

Definitely relevant for 'era relative' / 'absolute' arguments. But of course you can only calculate standard deviations if you have full league stats which I don't have ofc. Could be interesting to ask Thinking Basketball about this though.

Lower and upper-bounds are a useful concept too I think. Whatever you think of 94, 1995 drop is probably underrating Jordan's cast, while the 71 Celtics drop relative to 69 probably overrates Russell's.
2010-11 LeBron’s Cavaliers: http://20secondtimeout.blogspot.com/2011/07/analyzing-collapse-of-2010-11-cleveland.html?m=1 / [url]ESPNwww.espn.comCleveland Cavaliers: Why the Cavs have imploded in LeBron James' wake[/url]. Some
-Raw WOWY exaggerates the 2011 cavs drop because of LeBron, as it doesn’t include:

We have a 21-game sample from 2011 where the Cavs played the same starters Lebron was playing with. They posted a 18-win srs. Adjusted value being "+7" seems off. I would just use that 21-game sample.

Can do a lower-bound(likely undersells Lebron) adjustment with 2004 Lebron where you take boozer's net-rating from the next year and subtract it from the jump the 2004 cavs see with Lebron. Boozer was actually on the 2003 roster(and probably had improved by 2005) but iirc that lowers the jump to +4. Can check later

Huh, cool idea of an upper and lower bound based on different adjustments. I hadn't thought of that!

Re: 2011, yeah I struggled with that. I wanted to use the 210game sample when they were all healthy, but I already set the threshold for a reasonable WOWY sample at 30 games. It was basically an arbitrary threshold, but >33% of a season seems good, seems less likely to be too susceptible to a few blowouts or an extra difficult or easy schedule, other stats like RAPM start being a level less noisy as you go from 20 to 30 game samples so that's neat. We basically have to choose between having all players healthy or a 29% bigger sample size that's closer to 30 games with fewer healthy players. Either would be fine, if you prefer the 21-game sample you mention, sure thing.

One note, it sounds like from reports that those weren't playing fully healthy/fit in at least some of the games they did play. Not sure how to correct for this mathematically, but in theory that would under-sell the 11 Cavs 'healthy' sample, which might slightly overrate 2010 LeBron's WOWY. Regardless of what you choose, it's still clearly a Top 3 sample ever at worst. But this is the kind of stuff that makes a WOWY database hard... you can go deeper on so many samples, so it becomes a bit hard to have the most accurate / corrected values for everyone while being consistent.

1970 Kareem’s Bucks: https://forums.realgm.com/boards/viewtopic.php?p=104187436#p104187436, bottom of the post.
-WOWY underrates 1969 Bucks: Gained Flynn Robinson (3rd best player) and rookie Zaid Abdul-Aziz (5th in minutes) part way through season. Greg Smith (one of their best players) was a rookie
-WOWY overrates 1970 Bucks’ jump with Kareem: Bob Deandridge (3rd best player in 71) joined along with Kareem. Rookies Zaid Abdul-Aziz and Greg Smith likely improved too.
-Team context: Expansion tends to produce extra bad teams in their first years of existence, and the best teams tend to have overrated SRS because they feast on the bad teams. 1969 Bucks was the first year of existence (totally new team, first-time coach who was a player immediately prior). 1970 Bucks had higher SRS in expansion year.

Reverse applies to Kareem's Lakers. 75 roster loses players and SRS is generally lower as of 77 and beyond.

Fully agreed! That's why I wanted this to be a database we could update occasionally. It's hard for one person to make all the adjustments, but if you want to do the calculation (or if I get more time at some point in the next few weeks), this is just the kind of sample where we might add another 'Alternate value' that corrects for the context.

*Alternate Value: 1992–95 Bulls: Total Change: +5.26 [Alternate Value using 1992–93 for the ‘with’ sample, since many have argued Bulls were coasting in 93]
-1995–96 Bulls: 10.96 with, 4.29 without. Total change: +6.67 [re-joining Bulls]

95-96 doesn't really fit the rest of what is included here. Jordan was on the team. If you're looking fora 96 extrap, using 94 to 96 makes sense. If "player improves" is being included, then 70 -> 71 kareem, 08 -> 09 Lebron, 2014 -> 2015 steph all seem equally valid as inclusions

I'd also add Jordan is replaced with pete myers which is a pretty big downgrade but I tend to just use 92 anyway. Am curious how exactly you're adjusting for 1999.

I tried to keep it 2 years which is why I did 95–96. The *only* exception to the two years thing is the 92–95 Jordan, which I added as an Alternate sample just because of all the 92 discussion that inevitably follows that sample. I was hoping to preempt the discussion if I didn't include it :lol:

Out of curiosity, why would you think 94 vs 96 is better than 95 vs 96? The players and rosters I'd expect would be closer in 95 vs 96. The one thing I would have expected as a critique for that sample is not adjusting for the addition of Rodman... which is on the ever-growing list of samples that could be improved by adjusting for and important roster change that occurs at the same time.

Re: "player improves" WOWY, huh that's interesting. I'd keep it as a separate list. There's no real off sample... it's just a "with less-prime player" sample vs "with more-prime player" sample. Which still might have information! But might be a separate list from traditional 'WOWY'.

Re: 1998–99 adjustment for Jordan, the Adjusted Value formula is
A: 1998 Bulls MoV with Pippen playing - B: 1999 Bulls SRS - C: 1998 Pippen's WOWY - D: 1997 Rodman's WOWY.
A: Pippen missed half the year, so it seems unfair to use 1998 Bulls' full year SRS... which underrates them when healthy. Especially if we're correcting for Pippen.
B: Team SRS without Jordan/Pippen/Rodman. Pretty standard stuff.
C: Subtracting out 98 Pippen's WOWY from the change in team performance.
D: Subtracting out 97 Rodman's WOWY from the change in team performance. For Rodman and Pippen, I just used the closest raw WOWY score with a reasonable sample size, which happened to be the same year for Pippen and the year prior for Rodman. I'd expect 97 Rodman to slightly overrate 98 Rodman (so Jordan's Alternate Value WOWY would be slightly better if that were the only change)... but of course this isn't the only change (Phil Jackson changed teams), so I'm OK with overrating Rodman slightly if I don't account for the other changes.

DraymondGold · Post #9 » by **DraymondGold** » Tue Jul 25, 2023 3:43 am

homecourtloss wrote:
DraymondGold wrote:
homecourtloss wrote:
What are the numbers without any “adjusted” or “alternate” values?
Literally just raw WOWY, that's it. You can find an expanded definition with links to original sources under the "What is WOWY" spoiler in the OP.

But to be more specific: raw WOWY is just the change in team performance when a player plays in a game vs when they miss a game over a period of time. So for example, the formula for Raw WOWY is: Raw WOWY = [SRS (or Margin of Victory per game) when a player plays games for a certain team] - [SRS (or Margin of Victory per game) when a player misses games for a certain team].

Let's look at Michael Jordan, just to give a few examples.

Example 1: Michael Jordan Rookie Season sample. These are full year samples. Jordan played 0 games in 1984, and played 82 games in 1985. So we can use SRS to evaluate the change due to adding Jordan (and due to other roster changes, players aging, general variance, etc.).

So 1984–85 Jordan's raw WOWY = [SRS in season when he played] - [SRS in season when he didn't] = [1985 Bulls' SRS of -0.50] – [1984 Bulls' SRS of -4.69 = +4.19

Example 2: Michael Jordan's 1986 Injury. right off the bat, this sample is different. Jordan didn’t miss the full year, he missed only part of the year (missed 64 games, played 18). Like I said in OP, I’m using Margin of Victory per game for sub-season samples. So we get MoV for 1986 Bulls in games with Jordan and without Jordan.

But we reach a second problem: I set a minimum threshold of 30 games with and without, while we only have 18 games with. So we need to go to a neighboring year. Since Jordan was playing at the end of the 1986 season, then 1987 is closer for a “with” sample, so let’s go with 1987 (which has the added benefit of being a non-rookie year, which seems fairer to Jordan). We get the SRS or MoV for 1987 Bulls in games with and without Jordan (in this case, he played the full season so we get the team SRS… if he missed a non-negligible number of games we’d get the “with Jordan” MoV and “without Jordan” MoV).

Then we calculate the WOWY: 1986-87 WOWY = [average 86-87 Bulls MoV or SRS in games with Jordan playing] - [average 86-87 Bulls MoV or SRS in games without Jordan playing] = [0.38 with] - [-3.86 without] = 4.24.

But I made a somewhat arbitrary choice here. I chose to use 1987 for the second sample, rather than the other neighboring year (1985). I thought it was a better choice, but it was still arbitrary. So we can do the same process using 1985 instead of 1986 to get the “Alternate Value”. I include the alternate value for completeness.

Example 3: Michael Jordan's 1998 Retirement.
On the face of it, this is like Example 1. Jordan played the entire 1998 year and missed the entire 1999 year, so we can sue SRS for both. And that’s what I do for the standard value.

So why do we have an “Alternate Value” here? Well, there were pretty major other roster changes that we might correct for. Pippen and Rodman also left the Bulls at the same time Jordan retired. So this inspires the Alternate Value.

I check if there’s a raw WOWY sample for these two players within a year or two. I make sure it has a non-negligible sample size and doesn’t look like noise (e.g. it doesn’t put Pippen’s value at -10.0 which would obviously wrong). And as it turns out, there is!

Using the same process, we can get a raw WOWY for 1998 Pippen. We look at the 1998 Bulls MoV with and without Pippen, and find Pippen has a raw WOWY of +3.1. Rodman didn’t miss enough games in 98, but he did miss more in 97, and we find he has a raw WOWY of +2.75. So to isolate Jordan’s (approximate) WOWY, we take the full WOWY change in the 98 Bulls’ MoV with Pippen playing compared to the 99 Bulls’ SRS without any of them, then subtract Pippen and Rodman’s raw WOWY. The remainder is Jordan’s approximate value (or other contextual changes from the roster, coaching staff, etc).

So that’s the alternate value in 1998

…

If you have any other questions, feel free to check the article I linked in the “What is WOWY” spoiler above, or just ask if there’s any samples you’re curious about!

I am quite familiar with WOWY. Perhaps I wasn’t clear—I was asking for the career, 10 year, peak, etc., numbers that you listed for the players without the “alternate” and “adjusted” numbers.

Oh lol. Well I guess it's a useful explanation for any newbies.

Career and 10-year peak numbers are just averages (with everything equally weighted) of the samples included here. So if there's WOWY data with a sample size of <30 games missed (like the stuff you find in Thinking Basketball's full database), those aren't included.

When there's a standard value and an 'Alternate value', I picked the one that seemed most fair (but that's why I said you could make other choices).
A) So when I adjusted for other roster changes (e.g. 98 Jordan's alternate value that adjusts for Pippen or Rodman, 2008 KG's alternate value that adjusts for Ray Allen, etc.), I used those values for the career / prime average.
B) When the alternate values provided minor health corrections (e.g. 2019 LeBron's alternate value, which only uses games he played for the 'with' sample other than the full 2019 Lakers), I used those for the career / prime average.
C) But when there's an 'alternate value' that gives an alternate pair of years (e.g. do we use 1985–86 Jordan or 1986–76 Jordan for the 86 injury?)... that required an arbitrary choice. I tended to pick the one that made the player looked better (e.g. 1969–70 > 70–71 for Wilt, 1986–87 > 85–86 for Jordan, 88–89 > 89–90 for Bird, etc.).

I included the 10 years for their specific primes next to the players.

For the peaks, I was less careful... I literally just picked out the highest samples (or in some cases the more accurate 'alternate values' e.g. the ones that adjust for 98-99 Pippen/Rodman) in the database.

Post #10 » by **Moonbeam** » Tue Jul 25, 2023 6:44 am

Thanks for sharing all of this! I've also been looking at multi-year WOWY samples from box scores and have computed my own versions. I'll have to look into how Thinking Basketball does their WOWYR as my stuff is purportedly similar to it.

One thing that jumps out at me initially is that when I've looked at 5-year windows, Magic consistently finishes ahead of Bird throughout Magic's career, but he falls behind here.

ShaqAttac · Post #11 » by **ShaqAttac** » Tue Jul 25, 2023 7:17 am

DraymondGold wrote:Michael Jordan
-1984–85 Bulls: -0.5 with, -4.69 without. Total change: +4.19 [Rookie year]
-1986–87 Bulls: 0.38 with, -3.86 without. Total change: +4.24 [Injury year]
*Alternate Value: 1985–86 Bulls: Total change: +2.8 [Alternate Value uses 1985 instead of 1987]
-1993–94 Bulls: 6.19 with, 2.87 without. Total change: +3.32 [Retirement]
*Alternate Value: 1992–95 Bulls: Total Change: +5.26 [Alternate Value using 1992–93 for the ‘with’ sample, since many have argued Bulls were coasting in 93]
-1995–96 Bulls: 10.96 with, 4.29 without. Total change: +6.67 [re-joining Bulls]
-1998–99 Bulls: 7.24 with, -8.58 without. Total change: +15.82 [Retirement]
*Alternate Value: 1998–99 Bulls: 8.55 with, -8.58 without. Total change: +11.28 [Alternate Value uses MoV with Pippen for ‘with’ sample, then subtract’s 98 Pippen’s 3.1 WOWY and 97 Rodman’s 2.75 WOWY].
-2001–02 Wizards: -1.57 with, -6.75 without. Total change: +5.18 [joining Wizards]
*Alternate Value: 2001–02 Wizards: Total change: +5.51 [Alternate Value only uses games Jordan played for ‘with’ sample]/
-2003–04 Wizards -1.47 with, -6.12 without. Total change: +4.65 [Retirement]
Career Average: +5.69 (using latter 2 alternate values)
10-year prime: +7.09 (1989–1998, +7.74 1989–1998 using alternate value for 1993 too)
Non-prime average: +4.65

uhhhhhhhhh, mj was on the team in 95. also u boostin 86 by usin 87. do u do that for other guys?

anyway ig by this list bron is tops. curry 2nd so maybe i should vote him higher? russ the real top coz wins worth more in 60s

cap sucks here for some reason. in the stuff kd, em, n 70s posted he looked 2nd after bron. why he so much lower here?

SHAQ32 · Post #12 » by **SHAQ32** » Tue Jul 25, 2023 7:27 am

Put some respek on Bird's name!

DraymondGold · Post #13 » by **DraymondGold** » Tue Jul 25, 2023 3:08 pm

ShaqAttac wrote:
DraymondGold wrote:Michael Jordan
-1984–85 Bulls: -0.5 with, -4.69 without. Total change: +4.19 [Rookie year]
-1986–87 Bulls: 0.38 with, -3.86 without. Total change: +4.24 [Injury year]
*Alternate Value: 1985–86 Bulls: Total change: +2.8 [Alternate Value uses 1985 instead of 1987]
-1993–94 Bulls: 6.19 with, 2.87 without. Total change: +3.32 [Retirement]
*Alternate Value: 1992–95 Bulls: Total Change: +5.26 [Alternate Value using 1992–93 for the ‘with’ sample, since many have argued Bulls were coasting in 93]
-1995–96 Bulls: 10.96 with, 4.29 without. Total change: +6.67 [re-joining Bulls]
-1998–99 Bulls: 7.24 with, -8.58 without. Total change: +15.82 [Retirement]
*Alternate Value: 1998–99 Bulls: 8.55 with, -8.58 without. Total change: +11.28 [Alternate Value uses MoV with Pippen for ‘with’ sample, then subtract’s 98 Pippen’s 3.1 WOWY and 97 Rodman’s 2.75 WOWY].
-2001–02 Wizards: -1.57 with, -6.75 without. Total change: +5.18 [joining Wizards]
*Alternate Value: 2001–02 Wizards: Total change: +5.51 [Alternate Value only uses games Jordan played for ‘with’ sample]/
-2003–04 Wizards -1.47 with, -6.12 without. Total change: +4.65 [Retirement]
Career Average: +5.69 (using latter 2 alternate values)
10-year prime: +7.09 (1989–1998, +7.74 1989–1998 using alternate value for 1993 too)
Non-prime average: +4.65

uhhhhhhhhh, mj was on the team in 95.

Not for the whole season, no. Jordan played 17 games and missed 65.

But remember, I set the threshold to be a 30 game ‘with’ sample and a 30 game so ‘without’ sample. So we have to look to the neighboring years to buff up the ‘with’ sample… so we go to 96, as he didn’t play at all in 94.

So the ‘with’ sample becomes [17 games with 95 Jordan + 82 games with 96 Jordan] and the without sample remains [65 games without 95 Jordan].

also u boostin 86 by usin 87. do u do that for other guys?

Yes I am. See Example C on the post above (post 9).

anyway ig by this list bron is tops. curry 2nd so maybe i should vote him higher? russ the real top coz wins worth more in 60s

I'd say LBJ has a clear case, but I wouldn't say it's 100% given bron is tops. Curry has best 10 year prime, then Bird (on only 2 samples), then LeBron, then Jordan and Russell. Oscar (2 samples), Bird (2 samples), Jordan, and Wilt (2 samples, basically equal to LeBron) have better non-10-year-prime years. In terms of peak samples, Bird has the best, then 98 Jordan (even after adjusting for Pippen/Rodman), then there's two LeBron samples. There's also 2 Curry samples above 8+ WOWY.

If you look at career value over total games played, I'd say it's probably LeBron. But it's not from getting separation in his 10 year prime or his non-prime or his peak samples... it's more from combining being *near* the top in all of those with having significantly more games than all the other players.

Curry's always been GOAT level prime in WOWY. I've always been confused as to why you n Ohayo weren't higher on Curry's prime given his WOWY.

And yep there's definitely an argument for Russell if you go to WOWY relative to era, rather than some sort of absolute WOWY. By relativity era, I mean looking at some sort of WOWY standard deviation (we might guess the raw WOWY back then was lower) or if we do longevity relative to era (because players played for less time) and thus boost Russell's longevity.

cap sucks here for some reason. in the stuff kd, em, n 70s posted he looked 2nd after bron. why he so much lower here?

You’d have to share the sample they’re using. My guess would be they’re focusing on 1970 sample or Kareem’s improvement (along with his teammates’ and Oscar) leading to team improvement in 1971.

But that’s the risk of using single samples, without looking at the broader range of samples. You risk both missing the better samples of other players, and you risk missing the worse samples of the player you're focusing on. WOWY is noisy and context/roster dependent, so I personally don’t like to over-index on any single sample.

In Thinking Basketball’s 10-year Prime WOWY (which only looks at singe-season WOWY averaged over a given time span, so no looking at off-season trades or changes with rookie years)… Kareem ranks 130th all time. So there’s plenty of raw WOWY samples that aren’t that high.

With that in mind, there’s a bunch of contextual reasons why 1970 is a slightly overrated WOWY sample (see the context part of OP).

But! I’d think the 130th all time ranking certainly underrates Kareem, as do some of the samples here. Why?
1) The mid-70s samples were traded where the non-Kareem team benefited from gaining a player. This is different from an injury year or joining in free agency. This will make Kareem’s ‘with’ sample look worse because Kareem’s not playing with someone the team used to have… and make the ‘without’ sample look better because his old team is now gaining a new player! But this is the risk of using WOWY… it doesn’t adjust for teammates. A better statistic would be Adjusted WOWY (like WOWYR) that adjusts for teammates, or we cold try to approximate adjustments by subtracting out those player’s WOWYs if we can get them.
2) The only old Kareem sample we have is his retirement. He was far worse as a 40 year old in the late 80s than he was in his late prime / early post-prime in 80–85. So the unequal sampling of missed games happens to only get to non-prime Kareem when he was at his worst.

My guess is it's somewhere in the middle. Kareem's not as good as the 70 sample alone suggests, but he's higher than the mid-70s samples suggest. And that's exactly what the adjust WOWY metrics show

DraymondGold · Post #14 » by **DraymondGold** » Tue Jul 25, 2023 3:21 pm

Moonbeam wrote:Thanks for sharing all of this! I've also been looking at multi-year WOWY samples from box scores and have computed my own versions. I'll have to look into how Thinking Basketball does their WOWYR as my stuff is purportedly similar to it.

One thing that jumps out at me initially is that when I've looked at 5-year windows, Magic consistently finishes ahead of Bird throughout Magic's career, but he falls behind here.

I've loved reading your posts in the other thread!

Definitely looking forward to seeing the full dataset when it's done.

Agreed, the Magic Bird comparison is super interesting here. Note that I'm only using larger samples (30+ on / off games) and that I am including WOWY across multiple years (e.g. including off-season trades, rookie's joining in the off-season, etc.)... which Thinking Basketball does not.

One thing that stuck out to me with the 5 year thing... Thinking Basketball mentioned that Bird is actually higher than Magic in his adjusted WOWY metrics in the early years (he did a ridge regression for 1954-1983 to look at the older guys first: https://thinkingbasketball.net/2016/09/28/iii-historical-impact-wowyr-60-years-of-plus-minus/). Presumably some difference about the years (4 years from 80–83 rather than 5) or differences in the regression account for TB having Bird higher early on while your numbers have Magic higher?

Another point: TB mentioned that Bird's WOWYR error is higher among all-time players ("Pretty much any tweaking of variables yields the same names at the top (although of the big names, Larry Bird moves around a bit", https://thinkingbasketball.net/2016/09/28/iii-historical-impact-wowyr-60-years-of-plus-minus/) and that some of this uncertainty is due to Reggie Lewis happening to join the team at high minutes right when the Celtics were at their best ("Bird has lots of instability in his result because of Reggie Lewis", https://thinkingbasketball.net/2017/11/17/part-iv-historical-impact-multiple-wowyr-studies/).

I'd be interested to see if there's a similar thing happening with Bird here!

OhayoKD · Post #15 » by **OhayoKD** » Tue Jul 25, 2023 4:14 pm

So i think the sample treshold is causing alot of the unexpected results here:

ShaqAttac wrote:
DraymondGold wrote:
uhhhhhhhhh, mj was on the team in 95. also u boostin 86 by usin 87. do u do that for other guys?

anyway ig by this list bron is tops. curry 2nd so maybe i should vote him higher? russ the real top coz wins worth more in 60s

cap sucks here for some reason. in the stuff kd, em, n 70s posted he looked 2nd after bron. why he so much lower here?

Moonbeam wrote:Thanks for sharing all of this! I've also been looking at multi-year WOWY samples from box scores and have computed my own versions. I'll have to look into how Thinking Basketball does their WOWYR as my stuff is purportedly similar to it.

One thing that jumps out at me initially is that when I've looked at 5-year windows, Magic consistently finishes ahead of Bird throughout Magic's career, but he falls behind here.

The biggest difference between Dray's database and what you might find if you look at what I and others have provided(ben, 70sfan, eminence, ect) is we're using a lower filter. When you set it to 30 games the vast majority of samples get thrown out. So Kareem's 1975, Hakeem's 88, Lebron's 2015, Bird's 87/88 get thrown and alot of weight starts getting put on a specific year.

Bird's sample now is mostly 1980, Kareem loses his best wowy bit(1977/1971/1972/1974 do not have "withouts" unless you extrapolate somewhat), Magic loses 88, ect, so you get very different results than if more samples were being included.

There's also the matter of dray partially compensating for this by mixing different years for on/off. I'm guessing Jordan is the biggest benefactor here(87 and 96 are probably improved casts from the previous year and the team's on is higher).

You also have the matter of adjustments like 1999 where the team is stripped of everything really but without a way to quanitfy coaching or whatever intangible effects losing your three best players at once might have, you're left with what's probably not an indicative off-sample.

Adjustments themselves are also filtered. with 30-games we can't use the Bucks oscar-less games in 1972 which would likely benefit Kareem's other scores significantly.

I use 10-games as my filter allowing for smaller samples as corroborative signals so you get alot more data. Ben seems to be willing to go lower than that if nothing's available, but he seems to stick within that range. As does eminence and 70's fan.

DraymondGold wrote:
OhayoKD wrote:Appreicate the effort here. A few notes.

Depends on what you're using. If you are using "indirect" you get the largest possible samples in terms of off. Rest seems fine. Various adjustments like "full-strenght" and the like can be used to reduce noise.
The indirect stuff definitely helps give a larger off value, and 'full strength' adjustments definitely help reduce noise (though not entirely). So agreed there

On/off is still a larger sample regardless. To be more explicit, on/off depends on possession samples and WOWY depends on game samples, and there are *far* more 'on' possessions and 'off' possessions than there are 'with a player' games and 'without a player' games. Stars pretty easily play over a thousand possessions in a season, and they're of course don't play a thousand games in one or two seasons (that would be a crazy season! )

Um...no? Counting with a bigger or smaller unit doesn't actually change the size of the sample. I can break 82 games into 82 times X possessions and that would still be a larger result.

For those with a "championship" focus, general league-wide context matters too. +4 in a league or era of +2's > +6 in a league or era of +8's(important with Russell, Wilt and Kareem specifically).
Sure. Basically this is saying there's raw WOWY and then raw WOWY standard deviations, right?

Definitely relevant for 'era relative' / 'absolute' arguments. But of course you can only calculate standard deviations if you have full league stats which I don't have ofc. Could be interesting to ask Thinking Basketball about this though.

Sans does standard deviation, but really, what matters is srs of the top-teams(or the teams most likely to threaten you from winning). In several years during Russell's times those teasm are at +2. In 1969 they're below +6. In 91 there's another +8 team and in 93 there's a bunch of +6 teams. I'm guessing there's a way to do a weighted sd that focuses on the top, but in certain cases(russell, kareem's laker years) where we see similar lift as other players where tresholds are higher, it's pretty clear they are advantaged with a "relative to what it takes to win a championship" approach.

Lower and upper-bounds are a useful concept too I think. Whatever you think of 94, 1995 drop is probably underrating Jordan's cast, while the 71 Celtics drop relative to 69 probably overrates Russell's.
2010-11 LeBron’s Cavaliers: http://20secondtimeout.blogspot.com/2011/07/analyzing-collapse-of-2010-11-cleveland.html?m=1 / [url]ESPNwww.espn.comCleveland Cavaliers: Why the Cavs have imploded in LeBron James' wake[/url]. Some
-Raw WOWY exaggerates the 2011 cavs drop because of LeBron, as it doesn’t include:

We have a 21-game sample from 2011 where the Cavs played the same starters Lebron was playing with. They posted a 18-win srs. Adjusted value being "+7" seems off. I would just use that 21-game sample.

Can do a lower-bound(likely undersells Lebron) adjustment with 2004 Lebron where you take boozer's net-rating from the next year and subtract it from the jump the 2004 cavs see with Lebron. Boozer was actually on the 2003 roster(and probably had improved by 2005) but iirc that lowers the jump to +4. Can check later
Huh, cool idea of an upper and lower bound based on different adjustments. I hadn't thought of that!

Have been making use of those in the project but as an example:

-> Give Jordan all the credit for 1988-1984(or alternatively 1986) Bulls delta, upper-bound(more likely to overrate), best teams are

-> Give Kareem 1977 - 1975, ignores trades, lower bound(more likely to underrates)

Kareem scores higher and the best teams post significantly lower srs that year so i give Kareem a pretty clear advantage in terms of lift

Re: 2011, yeah I struggled with that. I wanted to use the 210game sample when they were all healthy, but I already set the threshold for a reasonable WOWY sample at 30 games. It was basically an arbitrary threshold, but >33% of a season seems good, seems less likely to be too susceptible to a few blowouts or an extra difficult or easy schedule, other stats like RAPM start being a level less noisy as you go from 20 to 30 game samples so that's neat. We basically have to choose between having all players healthy or a 29% bigger sample size that's closer to 30 games with fewer healthy players. Either would be fine, if you prefer the 21-game sample you mention, sure thing.

Problem is there just aren't many 30-game samples lying around. I would probably set the threshold somewhere where you aren't forced to mix years for on/off. The less "adjustments" being made the better.

One note, it sounds like from reports that those weren't playing fully healthy/fit in at least some of the games they did play. Not sure how to correct for this mathematically, but in theory that would under-sell the 11 Cavs 'healthy' sample, which might slightly overrate 2010 LeBron's WOWY. Regardless of what you choose, it's still clearly a Top 3 sample ever at worst. But this is the kind of stuff that makes a WOWY database hard... you can go deeper on so many samples, so it becomes a bit hard to have the most accurate / corrected values for everyone while being consistent.

I would not know. I'm just using what Ben said. Worth considering if true.

Out of curiosity, why would you think 94 vs 96 is better than 95 vs 96? The players and rosters I'd expect would be closer in 95 vs 96. The one thing I would have expected as a critique for that sample is not adjusting for the addition of Rodman... which is on the ever-growing list of samples that could be improved by adjusting for and important roster change that occurs at the same time.

Because Rodman replaces Grant and Jordan is not on the team.

Re: "player improves" WOWY, huh that's interesting. I'd keep it as a separate list. There's no real off sample... it's just a "with less-prime player" sample vs "with more-prime player" sample. Which still might have information! But might be a separate list from traditional 'WOWY'.

I mean that is basically what you are doing with 95/96 and 86/87.

I'd say LBJ has a clear case, but I wouldn't say it's 100% given bron is tops. Curry has best 10 year prime, then Bird (on only 2 samples), then LeBron, then Jordan and Russell. Oscar (2 samples), Bird (2 samples), Jordan, and Wilt (2 samples, basically equal to LeBron) have better non-10-year-prime years. In terms of peak samples, Bird has the best, then 98 Jordan (even after adjusting for Pippen/Rodman), then there's two LeBron samples. There's also 2 Curry samples above 8+ WOWY.

If you look at career value over total games played, I'd say it's probably LeBron. But it's not from getting separation in his 10 year prime or his non-prime or his peak samples... it's more from combining being *near* the top in all of those with having significantly more games than all the other players.

Well there is also the career average where Lebron scores ahead of everyone but Bird despite having the longest career. Granted Russell's +4.9 is probably better with era-considerations but it speaks to those two having the most bullet-proof impact portfolios.

I have Kareem similar to Lebron but him falling this low with a few unfavorable filters/adjustments speaks to the uncertainity there. But Lebron's score could very easily go significantly higher with a lower filter or more favorable adjustments while Bird and Jordan are pretty close to being dialed up as high as possible.

lessthanjake · Post #16 » by **lessthanjake** » Tue Jul 25, 2023 6:10 pm

I’m very appreciative to have this database.

I kind of think the main conclusion I get from looking at this, though, is that WOWY is just incredibly noisy, to the point where I’m not sure how much value it has—particularly when comparing what happens when a player leaves or joins a team.

I think we can see that with the differences in WOWY for the same players when measuring the team they left and the team they joined. Some examples:

- The Royals only got 0.41 worse when Oscar left, but his presence simultaneously correlated with the Bucks getting 7.66 better. Which would on one end suggest Oscar was not very impactful at all at that point, while simultaneously indicating that he was very impactful.
- Old Hakeem leaves the Rockets and they get 7.02 worse, but then the team he joined also gets 2.39 worse. Which would suggest that Hakeem at the same point was both very positively impactful and also a negative impact.
- Shaq leaves the Magic and they get 5.47 worse, but then the team he joins gets 0.55 worse too. Again, this suggests that Shaq at the same time was significantly positively impactful, and also a negative impact.
- Shaq leaves the Heat and they get 9.26 worse, but then the team he joins also gets 2.62 worse. Again, completely opposite indicators about a player at the same point in their career.
- Garnett leaves the Wolves, and they only get 3.1 worse, but the Celtics get 9.30 better with him, even after adjusting for Ray Allen. So 2007-2008 Garnett looks modestly impactful on one side of things and hugely impactful on the other.
- Garnett leaves the Celtics and they get 4.35 worse, but then joins the Nets and they get 2.82 worse too—again, suggesting he was simultaneously both pretty impactful and a negative.
- Wilt left the 1969 76ers and they got 3.15 worse, but then the team he joined also got 1.15 worse. Again, modestly impactful and a negative at the same time.
- LeBron leaves the Cavs and they get 15.05 worse (or 10.94 adjusting for Shaq), but then the Heat only get 4.77 better (and that’s without a Bosh adjustment). So, hugely impactful at the same time as merely pretty impactful.
- LeBron leaves the Cavs and they get 9.98 worse, at the same time as he joins the Lakers and they only get 0.11 better (0r 1.09 better with a health adjustment). So LeBron is suggested to have been both hugely impactful and of little impact whatsoever.

Granted, there are a few examples where a player changes teams and the impacts look similar. For instance, Shaq leaving the Lakers and joining the Heat had a similar effect on both teams under this measure. As did LeBron leaving the Heat and joining the Cavs. But, I think it’s actually pretty striking that the numbers in these scenarios are usually super different.

This amount of noise is perhaps not a surprise, because there’s so many confounding factors that go into this, especially when comparing when a player changes teams. To begin with, there’s just sheer randomness as a result of relatively low sample size—which all of these examples are, just as a consequence of what’s being measured. But there’s more than just that. A lot of these samples have substantial differences in the rosters themselves. Players’ form ebbs and flows from year to year—sometimes by a lot. There’s injuries and other reasons that other players missed games, and it’s impossible to account for all of that (though this database does try to roughly account for a small number of significant things).

Crucially, there’s also other factors inherent in the business of how basketball transactions are done—such that WOWY is dramatically affected by how a player left (i.e. trade vs. free agency; trade for current assets vs. trade for future assets, etc.). If a player is traded, then the team they’re going to is losing stuff and the team they’re leaving is gaining something—which will tend to lower the traded player’s WOWY, by different amounts depending on how good the traded players currently are. If a player is instead signed in free agency, then the team they’re leaving isn’t gaining anything, and the team they’re going to isn’t really losing much, so the situation is much more conducive to a high WOWY. Furthermore, by virtue of losing a big player in free agency, teams often immediately decide to shift to rebuilding (read: tanking), such that they aren’t even really trying to win anymore, which obviously has a big artificial effect on WOWY. Moreover, with this level of player, their free agency is something teams sometimes plan for beforehand, often stacking their rosters with players on expiring contracts and whatnot in preparation to make a free agency bid for that player. This will artificially make the team worse before the player gets there, since the team was not fully focusing on putting together a successful basketball roster as much as they were focusing on putting together a roster that’d give them cap space the next year. Similarly, in many cases, the team that a free agent leaves has structured their roster with older veterans that’ll be solid players while the impending free agent is still there (that big player typically wants to win now, and the team knows they have a narrow window and are just trying to maximize that specific window), but those sorts of guys deteriorate quickly.

These factors are all really significant, and the fact that there’s very little correlation between a player’s WOWY regarding the team they leave and their WOWY regarding the team they join suggests to me that these confounding factors probably just swallow the entire inquiry with regards to WOWY between different seasons. And they also very likely systematically overrate players who left in free agency, particularly highly anticipated free agency.

scrabbarista · Post #17 » by **scrabbarista** » Tue Jul 25, 2023 6:28 pm

Anything that has '98 as MJ's peak is sus for me. That was like his ninth-best season. 2010 for LBJ is also strange. Not in his five best seasons, at the least.

Bklynborn682 · Post #18 » by **Bklynborn682** » Tue Jul 25, 2023 6:40 pm

Did the lakers truly get -.55 worse in 97 with the addition of shaq?

DraymondGold · Post #19 » by **DraymondGold** » Tue Jul 25, 2023 6:54 pm

OhayoKD wrote:So i think the sample treshold is causing alot of the unexpected results here:
ShaqAttac wrote:
DraymondGold wrote:
uhhhhhhhhh, mj was on the team in 95. also u boostin 86 by usin 87. do u do that for other guys?

anyway ig by this list bron is tops. curry 2nd so maybe i should vote him higher? russ the real top coz wins worth more in 60s

cap sucks here for some reason. in the stuff kd, em, n 70s posted he looked 2nd after bron. why he so much lower here?

Moonbeam wrote:Thanks for sharing all of this! I've also been looking at multi-year WOWY samples from box scores and have computed my own versions. I'll have to look into how Thinking Basketball does their WOWYR as my stuff is purportedly similar to it.

One thing that jumps out at me initially is that when I've looked at 5-year windows, Magic consistently finishes ahead of Bird throughout Magic's career, but he falls behind here.

The biggest difference between Dray's database and what you might find if you look at what I and others have provided(ben, 70sfan, eminence, ect) is we're using a lower filter. When you set it to 30 games the vast majority of samples get thrown out. So Kareem's 1975, Hakeem's 88, Lebron's 2015, Bird's 87/88 get thrown and alot of weight starts getting put on a specific year.

Bird's sample now is mostly 1980, Kareem loses his best wowy bit(1977/1971/1972/1974 do not have "withouts" unless you extrapolate somewhat), Magic loses 88, ect, so you get very different results than if more samples were being included.

…

I use 10-games as my filter allowing for smaller samples as corroborative signals so you get alot more data. Ben seems to be willing to go lower than that if nothing's available, but he seems to stick within that range. As does eminence and 70's fan.

Yep! To reiterate something I tried to say in the OP: these 10-year Prime or non-prime average or career average numbers are *not* supposed to be definitive! I included the player rankings because it’s the first question everyone would ask when comparing different WOWY data. These numbers likely underrate some players’ WOWY, properly rate some, and overrate others.

The goal of this database, in its current form (although it may grow over time

) is to get all the large-sample multi-year data into one place. We already have a database for the all the single-year data from Ben. But Ben doesn’t include multi-year data in his database! So this database just gets all that missing data in one spot. So for example, if someone asks “How does Garnett look in his multi-year WOWY data”… they can just look here, rather than having to manually calculate 10 samples themselves.

Re: player rankings using WOWY, I would definitely encourage including *every* available WOWY data. There’s no point in throwing out data if we have it. In this sense, Thinking Basketball’s career WOWY rankings are likely better, as it doesn’t throw out samples. But the limitation there is that Ben doesn’t include multi-year changes in his rankings… which helps limit large roster / context changes biasing the data, at the cost of a smaller sample size.

I’d recommend either using Ben’s Career / Prime rankings for now, or perhaps some eventual future evolution of this database that combine both the single-season WOWY data (of all sample sizes, from Ben’s database) along with the multi-year WOWY data (like rookie years, retirement years, off-season moves).

One note there: A proper prime average or career average would use intelligent averaging. You wouldn’t want to have every sample be equally weighted. The simplest way to do this would be to average with weights based on the limiting sample size, i.e. whichever is smaller between the ‘with’ sample and the ‘without’ sample, since that will be what dominates the noise. For example:

1) 1975 Kareem’s WOWY: +1.75 with (65 games), -5.71 without (17 games), Total change: +7.46.
But! 17 games is absolutely still noisy and can be biased by even 1 blowout. For example, in their worst game, the Bucks were blown out by 23 points by the Lakers (11/1/1974). This was a game where the Bucks didn’t yet have Jim Price (3rd in their mpg, joined half way through season). In fact Jim Price was actually playing for the other team! So the full-season-average Bucks are better than this one blowout suggests (as they hadn’t yet gotten one of their best players) and their full-season-average opponents are worse than this one blowout suggests (as they hadn’t yet lost one of their best players).
Removing just this *one game* from the sample improves the Bucks without-sample by +1.08 and brings 1975 Kareem’s WOWY from +7.46 to +6.38.

2) 1988 Hakeem’s WOWY: +1.63 with (79 games), -5.0 without (3 games). Total change: +6.63.
But! There’s just a 3 game off-sample for 1988 Hakeem (and only a 7 game off sample for 1987). Taking away even just one of these off-games reduces Hakeem’s WOWY by 2.5, all the way down to 4.13! And it’s worth noting that the Rockets roster was not constant in these without-games either. Ralph Sampson played 19 games this season, but he played the all with Hakeem (24% of Hakeems’ games). But he played none of three ‘without’ games this year, which hurts the Rockets more given he was their only other (actually minute-worth) Center.

To make this a little more specific: we can mathematically calculate our uncertainty based on the sample size, which is what Thinking Basketball did in the WOWY database. A sample size of 3 off-games has an 95% uncertainty bound off +/- 12.1 WOWY (i.e. we have 95% probability of being with 12.1 of the 'true WOWY', given a 3 game off sample). A sample size of 17 off games has a 95% uncertainty bound of +/- 5.2 WOWY. Compare that to a 30 game minimum sample where the uncertainty is +/- 2.9 WOWY. Note that these are calculated for single-season WOWY samples... over multi-year samples, the uncertainties may go up because we don't know a priori whether the other players will improve or get worse from year to year.

Now I’m *absolutely* not saying we should throw this WOWY data out. It’s good data, those games really did happen, etc. All I’m saying is that WOWY data is noisy, so I wouldn’t put too much faith into a single sample, particularly ones that have a small sample size, and particularly ones where there are other roster changes happening besides the player you’re interested in that may influence the team results. That’s why I’d suggest doing longer averages (e.g. 5 year or 10 year primes), where the averages are weighted by the sample size. That way you still include this data (which is positive for Kareem and Hakeem!) but don’t overrate it compared to larger samples, like the mid-70s trade for Kareem if you can adjust for the other roster changes or the 1986 injury for Hakeem.

Ben does do a more intelligent average for his primes, so that should be good. I haven’t checked recently, but I believe he uses the off-sample size as the weighting (since off is usually smallest?) then also accounts for uncertainty. He also accounts for home court vs away games, and for diminishing returns for having good WOWY on better teams. I’d have to look back to get the specifics right, but all that to say Ben uses a much more intelligent averaging system then I do here. He just doesn’t use the multi-year samples in his analysis, which may contain some information if you can correct for other roster changes.

OhayoKD wrote:There's also the matter of dray partially compensating for this by mixing different years for on/off. I'm guessing Jordan is the biggest benefactor here(87 and 96 are probably improved casts from the previous year and the team's on is higher).

You also have the matter of adjustments like 1999 where the team is stripped of everything really but without a way to quanitfy coaching or whatever intangible effects losing your three best players at once might have, you're left with what's probably not an indicative off-sample.

Adjustments themselves are also filtered. with 30-games we can't use the Bucks oscar-less games in 1972 which would likely benefit Kareem's other scores significantly.

For the multi-year data, this doesn’t seem too crazy to me in theory. If we’re going to compare a team’s change from before a player’s rookie year to during the rookie year (over a 2 year sample), it doesn’t seem crazy to consider a team’s change from during a player’s near-season-long-injury/mid-career-retirement to the next season after a player’s return from said near-season-long-injury/mid-career-retirement (over a 2 year sample).

Both are 2 year samples, with the ‘without’ sample predominantly in one year and the ‘with’ sample predominantly in the next. This obviously brings in the potential for other roster/context changes biasing the results… but that’s an inherent limit of raw WOWY. I’d always encourage proper evaluation of the context when using raw WOWY.

Re: Jordan, it's true that he's one of the benefactors by taking multi-year averages for 86–97 and 95–96 (along with e.g. 64 Wilt, others). The issue is the sample of 'with Jordan' games in 86 and 95 are small enough that the uncertainty is quite high compared to the true value. But yes, wrapping in other years does benefit him.
One note I’d add is that if you’re going to argue 98 overrates Jordan due to the loss of a coach and the decline in intangibles from losing multiple players, there’s definitely other samples that would get dinged too. 2010 LeBron, for instance, since coach of the year Mike Brown left, as did the GM, and presumably the combination of Lebron James/Shaquille O'Neal/Zydrunas Ilgauskas/Delonte West caused an intangibles decline for the Cavs too. There are other players who experience coaching changes during the same timeframes we're looking at... Wilt for instance. Not disagreeing that 98 may overrate Jordan though… I’d agree with that! I’m just saying I’d be careful only dinging certain players by adding context, without applying equal context to the players you end up comparing him to.

Speaking of samples that are unexpectedly high… why in the world is 1980 Bird so high?? It looks like the Celtics lost Jo Jo White, who has a shockingly negative WOWY a few years earlier in 1978, at the same time they gained rookie Bird. Did they some how improve by losing players from the lineup? Was the Celtics coaching change in 1980 helping?

Re: Oscar-less bucks and Kareem, that’s one thing that WOWYR incorporates that WOWY doesn’t, which is one reason I tend to like WOWYR more in general. It allows us to incorporate other players’ data to get a better sense for the people we’re interested in!

OhayoKD wrote:
DraymondGold wrote: The indirect stuff definitely helps give a larger off value, and 'full strength' adjustments definitely help reduce noise (though not entirely). So agreed there

On/off is still a larger sample regardless. To be more explicit, on/off depends on possession samples and WOWY depends on game samples, and there are *far* more 'on' possessions and 'off' possessions than there are 'with a player' games and 'without a player' games. Stars pretty easily play over a thousand possessions in a season, and they're of course don't play a thousand games in one or two seasons (that would be a crazy season! )

Um...no? Counting with a bigger or smaller unit doesn't actually change the size of the sample. I can break 82 games into 82 times X possessions and that would still be a larger result.

Remember they're counting different things though!

A WOWY sample size of 1 has a 95% uncertainty of +/- 16.4 WOWY (per Thinking Basketball WOWY database). In other words, if you have a 1 game sample, you're 95% likely to be within 16.4 of the 'true' WOWY.

So when we're taking an average over a timespan (say a single year) of a bunch of noisy WOWY data, you're trying to find the "true mean". But the issue is you don't have enough data to confidently say the true mean, given how noisy each individual sample is. For example if you have only 17 missed games, your 95% confidence interval from this single sample is +/- 5.2. Which isn't great.

Compare that to +/- data, where we're not counting games, we're counting possessions. So we basically always have more than 10x the on sample and the off sample in a given season. So even if the uncertainty of a single +/- sample larger than the uncertainty of a single WOWY game sample, we have so many more possessions than we do games that our 95% uncertainty for the plus minus would be a lot smaller than +/- 5.2.

OhayoKD wrote:
Sure. Basically this is saying there's raw WOWY and then raw WOWY standard deviations, right?

Definitely relevant for 'era relative' / 'absolute' arguments. But of course you can only calculate standard deviations if you have full league stats which I don't have ofc. Could be interesting to ask Thinking Basketball about this though.

Sans does standard deviation, but really, what matters is srs of the top-teams(or the teams most likely to threaten you from winning). In several years during Russell's times those teasm are at +2. In 1969 they're below +6. In 91 there's another +8 team and in 93 there's a bunch of +6 teams. I'm guessing there's a way to do a weighted sd that focuses on the top, but in certain cases(russell, kareem's laker years) where we see similar lift as other players where tresholds are higher, it's pretty clear they are advantaged with a "relative to what it takes to win a championship" approach.

Lower and upper-bounds are a useful concept too I think. Whatever you think of 94, 1995 drop is probably underrating Jordan's cast, while the 71 Celtics drop relative to 69 probably overrates Russell's.

We have a 21-game sample from 2011 where the Cavs played the same starters Lebron was playing with. They posted a 18-win srs. Adjusted value being "+7" seems off. I would just use that 21-game sample.

Can do a lower-bound(likely undersells Lebron) adjustment with 2004 Lebron where you take boozer's net-rating from the next year and subtract it from the jump the 2004 cavs see with Lebron. Boozer was actually on the 2003 roster(and probably had improved by 2005) but iirc that lowers the jump to +4. Can check later
Huh, cool idea of an upper and lower bound based on different adjustments. I hadn't thought of that!

Have been making use of those in the project but as an example:

-> Give Jordan all the credit for 1988-1984(or alternatively 1986) Bulls delta, upper-bound(more likely to overrate), best teams are

-> Give Kareem 1977 - 1975, ignores trades, lower bound(more likely to underrates)

Kareem scores higher and the best teams post significantly lower srs that year so i give Kareem a pretty clear advantage in terms of lift

Interesting stuff! The thing to be cautious of here is that even without trades, all the other players are growing, with some getting better and some worsening. Our uncertainties will go up because of this.

Re: 2011, yeah I struggled with that. I wanted to use the 210game sample when they were all healthy, but I already set the threshold for a reasonable WOWY sample at 30 games. It was basically an arbitrary threshold, but >33% of a season seems good, seems less likely to be too susceptible to a few blowouts or an extra difficult or easy schedule, other stats like RAPM start being a level less noisy as you go from 20 to 30 game samples so that's neat. We basically have to choose between having all players healthy or a 29% bigger sample size that's closer to 30 games with fewer healthy players. Either would be fine, if you prefer the 21-game sample you mention, sure thing.

Problem is there just aren't many 30-game samples lying around. I would probably set the threshold somewhere where you aren't forced to mix years for on/off. The less "adjustments" being made the better.

It's a problem if your focus is on getting a proper 10-year / career average! Which wasn't the focus of this database per above. But yes, if we want to get better career averages, we should lower the threshold then weight things based on sample size or uncertainty or etc.

I'm a bit confused by your comment: "The less "adjustments" being made the better." Weren't you doing some adjustment with... was it Hakeem?... in the Top 100 project by subtracting out other major player's raw WOWY? I recall you saying this was in your opinion better than the raw WOWY, though I may be misremembering.

Unless your saying don't be inconsistent by adjusting for certain players / situations but not others?

If your point is raw WOWY in general is better than adjusted WOWY, I just disagree with that (but we've disagreed on this before, which is okay!). To me, in general, it's better to adjust for changes in rosters/teammates than it is to ignore them. APM and RAPM tell us more about the specific player's contribution than raw plus minus, although both tell us something. To me, the same is true for WOWY vs Adjusted WOWY, particularly when adjusted WOWY allows us to increase the sample size of data points by also bringing in the games from changing teammates and opponents. Of course, these adjustments still have uncertainty, and different adjustment / regression methods / priors can lead to different results. But to me at least, a priori, it's better to adjust for teammates than to just ignore them.

One note, it sounds like from reports that those weren't playing fully healthy/fit in at least some of the games they did play. Not sure how to correct for this mathematically, but in theory that would under-sell the 11 Cavs 'healthy' sample, which might slightly overrate 2010 LeBron's WOWY. Regardless of what you choose, it's still clearly a Top 3 sample ever at worst. But this is the kind of stuff that makes a WOWY database hard... you can go deeper on so many samples, so it becomes a bit hard to have the most accurate / corrected values for everyone while being consistent.

I would not know. I'm just using what Ben said. Worth considering if true.

Out of curiosity, why would you think 94 vs 96 is better than 95 vs 96? The players and rosters I'd expect would be closer in 95 vs 96. The one thing I would have expected as a critique for that sample is not adjusting for the addition of Rodman... which is on the ever-growing list of samples that could be improved by adjusting for and important roster change that occurs at the same time.

Because Rodman replaces Grant and Jordan is not on the team.

Hmm... there's too many other changes in players improving/worsening for my taste, along with having more roster changes from 94 to 96 vs 95 to 96. 95 Pippen's closer in aging curve to 96 Pippen, same with Kerr, same with Kukoc, etc. Me personally, I'd rather 'adjust' for Rodman by subtracting out his raw WOWY from the 96 Bulls change, if we were to try to get our estimate of 95-96 Jordan's WOWY more accurate.

Re: "player improves" WOWY, huh that's interesting. I'd keep it as a separate list. There's no real off sample... it's just a "with less-prime player" sample vs "with more-prime player" sample. Which still might have information! But might be a separate list from traditional 'WOWY'.

I mean that is basically what you are doing with 95/96 and 86/87.
I'd say LBJ has a clear case, but I wouldn't say it's 100% given bron is tops. Curry has best 10 year prime, then Bird (on only 2 samples), then LeBron, then Jordan and Russell. Oscar (2 samples), Bird (2 samples), Jordan, and Wilt (2 samples, basically equal to LeBron) have better non-10-year-prime years. In terms of peak samples, Bird has the best, then 98 Jordan (even after adjusting for Pippen/Rodman), then there's two LeBron samples. There's also 2 Curry samples above 8+ WOWY.

If you look at career value over total games played, I'd say it's probably LeBron. But it's not from getting separation in his 10 year prime or his non-prime or his peak samples... it's more from combining being *near* the top in all of those with having significantly more games than all the other players.

Well there is also the career average where Lebron scores ahead of everyone but Bird despite having the longest career. Granted Russell's +4.9 is probably better with era-considerations but it speaks to those two having the most bullet-proof impact portfolios.

I have Kareem similar to Lebron but him falling this low with a few unfavorable filters/adjustments speaks to the uncertainity there. But Lebron's score could very easily go significantly higher with a lower filter or more favorable adjustments while Bird and Jordan are pretty close to being dialed up as high as possible.

I'd be careful just going with career averages, given how unequal the sampling is (e.g. some players like Duncan really just have rookie/retirement samples without much prime, others have tons of prime samples). But yes, LBJ looks great in raw WOWY.

I'm higher on prime Curry's raw WOWY... it just looks consistently GOAT level in every sample I look at. I'm a bit puzzled why you're not as high on Curry, given how much you value WOWY and how great he looks there.

Me personally, Not quite as high on Kareem's raw WOWY: the overall single-season data average is quite low per Ben's database, the multi-season stuff has some great samples like 70 at 75 but 70 is context-boosted and there are some lower samples too like the ones I listed, although those are likely underrating him given the trade.

OhayoKD · Post #20 » by **OhayoKD** » Tue Jul 25, 2023 7:30 pm

scrabbarista wrote:Anything that has '98 as MJ's peak is sus for me. That was like his ninth-best season. 2010 for LBJ is also strange. Not in his five best seasons, at the least.

The "peaks" are largely not included here. 1998 is a year where the Bulls lose everything and predictably outperforms "prime" signals for Mike like 1993 or the "upper-bound" for 1988. Considering 2010 comes right after 2009 which is pretty definitively the best statistical season in the last 40 years(and likely better than the box when you account for annoying things like defense, defensive attention, creative effeciency, and being an on-court general), I'm not really sure why we'd be questioning 2010 looking high.

That seems more like a matter of you letting priors bias your weighting of the data. Lebron's 09 and 10 are outliers and if you are under the impression those were not his best years, everything that follows should be curved up accordingly.