Who Are The Most Statistically Impactful Players in NBA History? (WOWY Type Discussion)

[ty 4191]

I was on the "Is Steph Curry Still Unappreciated" Thread and posted this:

As incredibly hyped as he's been, I still think he's perhaps still underestimated (in an ATG/historical sense):

All stats include the playoffs:

--Warriors since 2009-2010 without Curry playing: 76-134 (.362)
--Warriors since 2009-2010 with Curry playing: 643-328 (.662)

I challenge anyone to find a player with a really long career that has a bigger differential.

I was strongly encouraged to start a discussion just about this topic. I think it bears discussion.

So, who are the (other) most impactful players of all time, besides Curry, in terms of team winning percentage on/off or "WOWY"?

[RCM88x]

Quickly did Lebron thru 2022

54-102 without (.346)
894-472 with (.654)

Almost identical haha

If you adjust to when he stopped playing on his rookie contract (2008-2022) those numbers are quite different interestingly

45-99 without (.312)
726-324 with (.691)

[DraymondGold]

I was on the "Is Steph Curry Still Unappreciated" Thread and posted this:

As incredibly hyped as he's been, I still think he's perhaps still underestimated (in an ATG/historical sense):

All stats include the playoffs:

--Warriors since 2009-2010 without Curry playing: 76-134 (.362)
--Warriors since 2009-2010 with Curry playing: 643-328 (.662)

I challenge anyone to find a player with a really long career that has a bigger differential.

I was strongly encouraged to start a discussion just about this topic. I think it bears discussion.

So, who are the (other) most impactful players of all time, besides Curry, in terms of team winning percentage on/off or "WOWY"?

Huh, cool stat ty! I did a similar WOWY estimation for Curry in the Greatest Peaks project (https://forums.realgm.com/boards/viewtopic.php?p=100487575#p100487575)

It turns out there's different types of WOWY stats, just like there's different types of plus minus stats. It looks like you looked at how the record changes with and without a player -- but you could also look at how a team's point differential changes in games when a player plays vs when they don't. This is sort of like on-off stats, but here we're using margin of victory during a player's Games Played for the "on" sample and margin of victory during Games Injured/Rested as the "off" sample (rather than possessions played for the "on" sample and possessions rested for the "off" sample). In the link I posted above, I did an estimate (with a bit more uncertainty) for how Curry ranks all time, and he ends up 1st all time in this version of WOWY

The biggest issue is how to handle 2020; since Klay was also out for that year, the Warriors performed worse than if just Steph was out. So the two (lazy) options are to include 2020 but try to account for the fact that Klay was missing, or use the other non-2020 years. (or you could do a full regression to get the actual number, but that's a pain, so I'll just wait for Thinking Basketball to post it eventually).

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see. The biggest trends I've noticed are that WOWY tends to rank playmakers very highly (over play finishers / iso scorers), as well as defensive anchors (over perimeter defenders). My guess is because playmakers and defensive anchors are harder to replace in a night-to-night gameplan, they end up being missed more in the games they miss.

[migya]

Robinson must be up there.

[OhayoKD]

I was on the "Is Steph Curry Still Unappreciated" Thread and posted this:

As incredibly hyped as he's been, I still think he's perhaps still underestimated (in an ATG/historical sense):

All stats include the playoffs:

--Warriors since 2009-2010 without Curry playing: 76-134 (.362)
--Warriors since 2009-2010 with Curry playing: 643-328 (.662)

I challenge anyone to find a player with a really long career that has a bigger differential.

I was strongly encouraged to start a discussion just about this topic. I think it bears discussion.

So, who are the (other) most impactful players of all time, besides Curry, in terms of team winning percentage on/off or "WOWY"?

Huh, cool stat ty! I did a similar WOWY estimation for Curry in the Greatest Peaks project (https://forums.realgm.com/boards/viewtopic.php?p=100487575#p100487575)

It turns out there's different types of WOWY stats, just like there's different types of plus minus stats. It looks like you looked at how the record changes with and without a player -- but you could also look at how a team's point differential changes in games when a player plays vs when they don't. This is sort of like on-off stats, but here we're using margin of victory during a player's Games Played for the "on" sample and margin of victory during Games Injured/Rested as the "off" sample (rather than possessions played for the "on" sample and possessions rested for the "off" sample). In the link I posted above, I did an estimate (with a bit more uncertainty) for how Curry ranks all time, and he ends up 1st all time in this version of WOWY

The biggest issue is how to handle 2020; since Klay was also out for that year, the Warriors performed worse than if just Steph was out. So the two (lazy) options are to include 2020 but try to account for the fact that Klay was missing, or use the other non-2020 years. (or you could do a full regression to get the actual number, but that's a pain, so I'll just wait for Thinking Basketball to post it eventually).

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see. The biggest trends I've noticed are that WOWY tends to rank playmakers very highly (over play finishers / iso scorers), as well as defensive anchors (over perimeter defenders). My guess is because playmakers and defensive anchors are harder to replace in a night-to-night gameplan, they end up being missed more in the games they miss.

10 year averages are not how you want to use WOWY. It's not APM, you'll want to be specific about the samples you use

[penbeast0]

10 year averages are not how you want to use WOWY. It's not APM, you'll want to be specific about the samples you use

I don't think I understand. If you are going to use WOWY, what do you think is better than a 10 year average and why?

[OhayoKD]

10 year averages are not how you want to use WOWY. It's not APM, you'll want to be specific about the samples you use

I don't think I understand. If you are going to use WOWY, what do you think is better than a 10 year average and why?

Over 10 year spans or so, wowy will typically be based on a handful of games per season with players who've probably changed signifcantly during that time span. That's not really all that useful. What you want to do is look for moments where players or teammates miss(or are absent) from an unusally high number of games so you can get the largest samples. A good starting point for this would be the year before a player joins a team or the year after a leaves a team as you get a full season sample of data(70 Celtics, 84 Bulls, 84 Rockets, 69 Bucks, ect.) Then you can track roster changes, and granular stuff to adjust or guesstimate if the team improved got worse, ect. Lookign for concentrated stretches of missed game time, or how teams do when a star's teammates go out also can be useful. Ideally you want as much of this type of data for a player in various contexts and then you can compare players in these various situations directly.

Ex:

-> 69 Bucks are a 27 win team before they draft Kareem, win 56 games with rookie kareem and dandridge
-> 72 Bucks play at a 63 win pace in 18 games robertson misses
-> 75 Bucks go 3-14 in the 17 games Kareem misses, play at a 49 win pace with Kareem despite dandridge and robertson and falling off
-> 40 win difference with kareem and without when lakers miss the playoffs

In total this gives us an impact profile, on both high level teams(72) and low level teams(80, 75, 70) that's better than just about everyone in nba history, including players whose "average wowy" is higher(robinson). In fact, if we compare kareem and wowy leader robinson with like for like samples:

No, Kareem missing 17 games is the reason why. Do you think it's unreasonable explaination?

Not as much as the Spurs winning 59 in 96 with Robinson and only 20 in 97 without him.

38-19 With Kareem isn't great, not bad, but not elite winning.

Well, technically Robinson indeed had better WOWY than Kareem in these seasons:

1974/75 Bucks with Kareem: 35-30, 44 wins pace
1974/75 Bucks without Kareem: 3-14, 15 wins pace

1995/96 Spurs with Robinson: 59-23, 59 wins pace
1996/97 Spurs without Robinson: 20-52, 20 wins pace

The difference is that 1997 Spurs tanked hard to get Duncan in the draft, while Bucks tried to make the playoffs. I wouldn't compare these two situations at all.

A better comparison would be 1992 Spurs, when Robinson missed 14 games:

1991/92 Spurs with Robinson: 42-26, 51 wins pace
1991/92 Spurs without Robinson: 5-9, 30 wins pace

Now, you can see that Kareem actually played with worse team than Admiral. Besides, comparing him to Robinson isn't anything bad - in terms of RS lifting Robinson is more impressive than Jordan himself after all. The difference is that Kareem upped his production and impact in postseason, while Robinson collapsed.

Kareem also looks pretty good here, both as floor raiser and a cieling raiser relative to mamy people's preferred peak choice, Micheal Jordan who started with a 27 win team, and wasn't able to break 50 wins(despite a steadily improving supporting cast) until the implementation of the triangle in 90(upon which the bulls srs skyrocketed), and wasn't able to get past 55 wins, until improving offensive and defensive support(this is mostly Pippen becoming one of the best players in the league) had the bulls skyrocket from an average defense and good offense to the best defense and the best offense in the league.


If we just went by "prime wowy" Kareem wouldn't even score in the top 20. If we're more careful about how we use wowy, we get solid evidence for a player with a nigh unrivalled(lebron, hakeem, and russell are really the only players who don't look clearly worse via this type of analysis) influence at his apex who was able to sustain that influence on atg teams AND historic carry jobs across multiple contexts

[Dutchball97]

I was on the "Is Steph Curry Still Unappreciated" Thread and posted this:

As incredibly hyped as he's been, I still think he's perhaps still underestimated (in an ATG/historical sense):

All stats include the playoffs:

--Warriors since 2009-2010 without Curry playing: 76-134 (.362)
--Warriors since 2009-2010 with Curry playing: 643-328 (.662)

I challenge anyone to find a player with a really long career that has a bigger differential.

I was strongly encouraged to start a discussion just about this topic. I think it bears discussion.

So, who are the (other) most impactful players of all time, besides Curry, in terms of team winning percentage on/off or "WOWY"?

Huh, cool stat ty! I did a similar WOWY estimation for Curry in the Greatest Peaks project (https://forums.realgm.com/boards/viewtopic.php?p=100487575#p100487575)

It turns out there's different types of WOWY stats, just like there's different types of plus minus stats. It looks like you looked at how the record changes with and without a player -- but you could also look at how a team's point differential changes in games when a player plays vs when they don't. This is sort of like on-off stats, but here we're using margin of victory during a player's Games Played for the "on" sample and margin of victory during Games Injured/Rested as the "off" sample (rather than possessions played for the "on" sample and possessions rested for the "off" sample). In the link I posted above, I did an estimate (with a bit more uncertainty) for how Curry ranks all time, and he ends up 1st all time in this version of WOWY

The biggest issue is how to handle 2020; since Klay was also out for that year, the Warriors performed worse than if just Steph was out. So the two (lazy) options are to include 2020 but try to account for the fact that Klay was missing, or use the other non-2020 years. (or you could do a full regression to get the actual number, but that's a pain, so I'll just wait for Thinking Basketball to post it eventually).

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see. The biggest trends I've noticed are that WOWY tends to rank playmakers very highly (over play finishers / iso scorers), as well as defensive anchors (over perimeter defenders). My guess is because playmakers and defensive anchors are harder to replace in a night-to-night gameplan, they end up being missed more in the games they miss.

I think this possibly shows a flaw with building your entire defense or offense around one person because the moment they go to the bench it all falls apart. I do wonder how often this is by design (not investing much in a back-up because you want to play your main man as much as possible anyway and you still need to invest elsewhere too) or because of there simply being a shortage of good playmakers and defensive anchors available as back-ups. I feel like on/off stats could really benefit from the context of whether someone is breaking the stat because the team is built around them and their back-ups are terrible or if they're destroying the on/off stats despite having capable back-ups.

[AEnigma]

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see.

This is Ben’s (well, not strictly Ben’s, but he is the one who posted it) “Game-Level Adjusted Plus/Minus” (GPM) metric, not WOWY or even his own WOWYR (or its various iterations).
https://thinkingbasketball.net/tag/wowyr/

Purer “prime” WOWY is more like (Curry? >) Garnett > Lebron > Shaq > Nash > Bird > Hakeem > Magic > Robinson > Duncan > Kobe > Jordan. With the acknowledgment that the purer form is even more variable, it does still correlate reasonably well to longer RAPM and on/off studies, with Lebron, Garnett, and Shaq (and Curry) maintaining as the top ~ten-year prime performers since 1994/97.

[falcolombardi]

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see.

This is Ben’s (well, not strictly Ben’s, but he is the one who posted it) “Game-Level Adjusted Plus/Minus” (GPM) metric, not WOWY or even his own WOWYR (or its various iterations).
https://thinkingbasketball.net/tag/wowyr/

Purer “prime” WOWY is more like (Curry? >) Garnett > Lebron > Shaq > Nash > Bird > Hakeem > Magic > Robinson > Duncan > Kobe > Jordan. With the acknowledgment that the purer form is even more variable, it does still correlate reasonably well to longer RAPM and on/off studies, with Lebron, Garnett, and Shaq (and Curry) maintaining as the top ~ten-year prime performers since 1994/97.

I dont know of 10-year rapm but 5-year samples are that but replacing shaq with duncan (shaq, paul, wade right after) lebron usually a notch ahead

https://www.thespax.com/nba/quantifying-the-nbas-greatest-five-year-peaks-since-1997/

[falcolombardi]

10 year averages are not how you want to use WOWY. It's not APM, you'll want to be specific about the samples you use

I don't think I understand. If you are going to use WOWY, what do you think is better than a 10 year average and why?

Over 10 year spans or so, wowy will typically be based on a handful of games per season with players who've probably changed signifcantly during that time span. That's not really all that useful. What you want to do is look for moments where players or teammates miss(or are absent) from an unusally high number of games so you can get the largest samples. A good starting point for this would be the year before a player joins a team or the year after a leaves a team as you get a full season sample of data(70 Celtics, 84 Bulls, 84 Rockets, 69 Bucks, ect.) Then you can track roster changes, and granular stuff to adjust or guesstimate if the team improved got worse, ect. Lookign for concentrated stretches of missed game time, or how teams do when a star's teammates go out also can be useful. Ideally you want as much of this type of data for a player in various contexts and then you can compare players in these various situations directly.

Ex:

-> 69 Bucks are a 27 win team before they draft Kareem, win 56 games with rookie kareem and dandridge
-> 72 Bucks play at a 63 win pace in 18 games robertson misses
-> 75 Bucks go 3-14 in the 17 games Kareem misses, play at a 49 win pace with Kareem despite dandridge and robertson and falling off
-> 40 win difference with kareem and without when lakers miss the playoffs

In total this gives us an impact profile, on both high level teams(72) and low level teams(80, 75, 70) that's better than just about everyone in nba history, including players whose "average wowy" is higher(robinson). In fact, if we compare kareem and wowy leader robinson with like for like samples:

Not as much as the Spurs winning 59 in 96 with Robinson and only 20 in 97 without him.

38-19 With Kareem isn't great, not bad, but not elite winning.

Well, technically Robinson indeed had better WOWY than Kareem in these seasons:

1974/75 Bucks with Kareem: 35-30, 44 wins pace
1974/75 Bucks without Kareem: 3-14, 15 wins pace

1995/96 Spurs with Robinson: 59-23, 59 wins pace
1996/97 Spurs without Robinson: 20-52, 20 wins pace

The difference is that 1997 Spurs tanked hard to get Duncan in the draft, while Bucks tried to make the playoffs. I wouldn't compare these two situations at all.

A better comparison would be 1992 Spurs, when Robinson missed 14 games:

1991/92 Spurs with Robinson: 42-26, 51 wins pace
1991/92 Spurs without Robinson: 5-9, 30 wins pace

Now, you can see that Kareem actually played with worse team than Admiral. Besides, comparing him to Robinson isn't anything bad - in terms of RS lifting Robinson is more impressive than Jordan himself after all. The difference is that Kareem upped his production and impact in postseason, while Robinson collapsed.

Kareem also looks pretty good here, both as floor raiser and a cieling raiser relative to mamy people's preferred peak choice, Micheal Jordan who started with a 27 win team, and wasn't able to break 50 wins(despite a steadily improving supporting cast) until the implementation of the triangle in 90(upon which the bulls srs skyrocketed), and wasn't able to get past 55 wins, until improving offensive and defensive support(this is mostly Pippen becoming one of the best players in the league) had the bulls skyrocket from an average defense and good offense to the best defense and the best offense in the league.


If we just went by "prime wowy" Kareem wouldn't even score in the top 20. If we're more careful about how we use wowy, we get solid evidence for a player with a nigh unrivalled(lebron, hakeem, and russell are really the only players who don't look clearly worse via this type of analysis) influence at his apex who was able to sustain that influence on atg teams AND historic carry jobs across multiple contexts

So in other words, wowy of random sparse games across multiple seasons is too noisy (for all you know most games you are getting are not meaningful and the "control" of the study -rival strenght and team quality- are too variable themselves)

Does this mean the best use of wowy is as an "before and after" thingh?

Like russel-less celtics going from champions to 30 wins in 1970? (With no other big loss in roster from what i remember, quite the opposite with havlicec improving all his numbers).

Edit: to go on this for how mindblowing that is

When jordan, pippen, rodman and phil jackson left the bulls in 99 the drop off was similar to russel alone leaving boston

Russel may have been the most valuable player in basketball history

[falcolombardi]

I was on the "Is Steph Curry Still Unappreciated" Thread and posted this:

As incredibly hyped as he's been, I still think he's perhaps still underestimated (in an ATG/historical sense):

All stats include the playoffs:

--Warriors since 2009-2010 without Curry playing: 76-134 (.362)
--Warriors since 2009-2010 with Curry playing: 643-328 (.662)

I challenge anyone to find a player with a really long career that has a bigger differential.

I was strongly encouraged to start a discussion just about this topic. I think it bears discussion.

So, who are the (other) most impactful players of all time, besides Curry, in terms of team winning percentage on/off or "WOWY"?

Huh, cool stat ty! I did a similar WOWY estimation for Curry in the Greatest Peaks project (https://forums.realgm.com/boards/viewtopic.php?p=100487575#p100487575)

It turns out there's different types of WOWY stats, just like there's different types of plus minus stats. It looks like you looked at how the record changes with and without a player -- but you could also look at how a team's point differential changes in games when a player plays vs when they don't. This is sort of like on-off stats, but here we're using margin of victory during a player's Games Played for the "on" sample and margin of victory during Games Injured/Rested as the "off" sample (rather than possessions played for the "on" sample and possessions rested for the "off" sample). In the link I posted above, I did an estimate (with a bit more uncertainty) for how Curry ranks all time, and he ends up 1st all time in this version of WOWY

The biggest issue is how to handle 2020; since Klay was also out for that year, the Warriors performed worse than if just Steph was out. So the two (lazy) options are to include 2020 but try to account for the fact that Klay was missing, or use the other non-2020 years. (or you could do a full regression to get the actual number, but that's a pain, so I'll just wait for Thinking Basketball to post it eventually).

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see. The biggest trends I've noticed are that WOWY tends to rank playmakers very highly (over play finishers / iso scorers), as well as defensive anchors (over perimeter defenders). My guess is because playmakers and defensive anchors are harder to replace in a night-to-night gameplan, they end up being missed more in the games they miss.

As others have pointed out 10 year samples can be incredibly noisy even if the amount of games is decent cause the "control" in the study (the teammates and rival quality) can wildly change fron a missed game one season to a missed game 8 years later.

Seems like focusing on single seasons with a big missed game sample (10+ at least imo) or full seasons "before and after" a player may be more useful

Example russel leaving the championship celtics in 70 (past his prime to boot) and the team falling to 30 wins. A drop off comparable to the 99 bulls losing pippen, jordan, rodman and phil jackson

[penbeast0]

Before is less favorable to Russell. The Celtics before drafting Russell were an above average NBA team and in the half season before they got him were off to an excellent start.

[OhayoKD]

Before is less favorable to Russell. The Celtics before drafting Russell were an above average NBA team and in the half season before they got him were off to an excellent start.

Didn't they steadily get worse after?

[Owly]

I don't think I understand. If you are going to use WOWY, what do you think is better than a 10 year average and why?

Over 10 year spans or so, wowy will typically be based on a handful of games per season with players who've probably changed signifcantly during that time span. That's not really all that useful. What you want to do is look for moments where players or teammates miss(or are absent) from an unusally high number of games so you can get the largest samples. A good starting point for this would be the year before a player joins a team or the year after a leaves a team as you get a full season sample of data(70 Celtics, 84 Bulls, 84 Rockets, 69 Bucks, ect.) Then you can track roster changes, and granular stuff to adjust or guesstimate if the team improved got worse, ect. Lookign for concentrated stretches of missed game time, or how teams do when a star's teammates go out also can be useful. Ideally you want as much of this type of data for a player in various contexts and then you can compare players in these various situations directly.

Ex:

-> 69 Bucks are a 27 win team before they draft Kareem, win 56 games with rookie kareem and dandridge
-> 72 Bucks play at a 63 win pace in 18 games robertson misses
-> 75 Bucks go 3-14 in the 17 games Kareem misses, play at a 49 win pace with Kareem despite dandridge and robertson and falling off
-> 40 win difference with kareem and without when lakers miss the playoffs

In total this gives us an impact profile, on both high level teams(72) and low level teams(80, 75, 70) that's better than just about everyone in nba history, including players whose "average wowy" is higher(robinson). In fact, if we compare kareem and wowy leader robinson with like for like samples:

Well, technically Robinson indeed had better WOWY than Kareem in these seasons:

1974/75 Bucks with Kareem: 35-30, 44 wins pace
1974/75 Bucks without Kareem: 3-14, 15 wins pace

1995/96 Spurs with Robinson: 59-23, 59 wins pace
1996/97 Spurs without Robinson: 20-52, 20 wins pace

The difference is that 1997 Spurs tanked hard to get Duncan in the draft, while Bucks tried to make the playoffs. I wouldn't compare these two situations at all.

A better comparison would be 1992 Spurs, when Robinson missed 14 games:

1991/92 Spurs with Robinson: 42-26, 51 wins pace
1991/92 Spurs without Robinson: 5-9, 30 wins pace

Now, you can see that Kareem actually played with worse team than Admiral. Besides, comparing him to Robinson isn't anything bad - in terms of RS lifting Robinson is more impressive than Jordan himself after all. The difference is that Kareem upped his production and impact in postseason, while Robinson collapsed.

Kareem also looks pretty good here, both as floor raiser and a cieling raiser relative to mamy people's preferred peak choice, Micheal Jordan who started with a 27 win team, and wasn't able to break 50 wins(despite a steadily improving supporting cast) until the implementation of the triangle in 90(upon which the bulls srs skyrocketed), and wasn't able to get past 55 wins, until improving offensive and defensive support(this is mostly Pippen becoming one of the best players in the league) had the bulls skyrocket from an average defense and good offense to the best defense and the best offense in the league.


If we just went by "prime wowy" Kareem wouldn't even score in the top 20. If we're more careful about how we use wowy, we get solid evidence for a player with a nigh unrivalled(lebron, hakeem, and russell are really the only players who don't look clearly worse via this type of analysis) influence at his apex who was able to sustain that influence on atg teams AND historic carry jobs across multiple contexts

So in other words, wowy of random sparse games across multiple seasons is too noisy (for all you know most games you are getting are not meaningful and the "control" of the study -rival strenght and team quality- are too variable themselves)

Does this mean the best use of wowy is as an "before and after" thingh?

Like russel-less celtics going from champions to 30 wins in 1970? (With no other big loss in roster from what i remember, quite the opposite with havlicec improving all his numbers).

Edit: to go on this for how mindblowing that is

When jordan, pippen, rodman and phil jackson left the bulls in 99 the drop off was similar to russel alone leaving boston

Russel may have been the most valuable player in basketball history

...
Sam Jones?

I guess as ever the lesson is the human memory is fallible.

Also fwiw Howell has a pretty dreadful year, partially age-related one assumes, though he bounces back the next year.

Then too on

When jordan, pippen, rodman and phil jackson left the bulls in 99 the drop off was similar to russel alone leaving boston

On the Chicago side, Pippen played 44 games that season, whilst Rodman was showing signs of being much diminished, lower-impact ... even MJ was slipping. To be fair Boston may not have maxed out Jones in the '69 regular season.

Then too Chicago suffer a -15.82 SRS move. Boston's, while large in absolute terms was a comparatively meager -6.95. For a more simplistic if familiar measure Boston's win% change is -0.17, Chicago's -0.496. They're both large, but not that similar. I will grant some limitations on using SRS with different size leagues where the team in question makes up a different proportion of forming the league average.

[Owly]

Before is less favorable to Russell. The Celtics before drafting Russell were an above average NBA team and in the half season before they got him were off to an excellent start.

Didn't they steadily get worse after?

Simple(-ish) version.

Better win% without Russell. (RS)
Better points dif with Russell. (RS)

confounding factors/context
Boston's very start is even stronger, then Sharman goes down and they slip back. He returns around Russell's arrival.
About a month after Russell's arrival, Boston's rotation quality is boosted by the arrival (from military service) of Frank Ramsey.
Boston's point diff in the playoffs was marginally better than either, though neither opponent had a positive SRS (Syracuse 6th of 8 -1.03; Hawks 4th, -0.27 though slightly better differential after Naulls for Martin trade).

IMO this is works against Russell as huge (probably works against even "large") impact on arrival, especially versus later very small sample WoWY stuff that makes him look elite.

Can dig out numbers if necessary.

[f4p]

Then too Chicago suffer a -15.82 SRS move. Boston's, while large in absolute terms was a comparatively meager -6.95. For a more simplistic if familiar measure Boston's win% change is -0.17, Chicago's -0.496. They're both large, but not that similar. I will grant some limitations on using SRS with different size leagues where the team in question makes up a different proportion of forming the league average.

yeah, the dropoffs aren't even remotely similar. the SRS really shows it. the bulls won 13 (out of 50) games with a predicted win total of 9, so the -0.496 win% is with the bulls winning damn near 50% more games than they should have.

[AEnigma]

Celtics: 1969 -> 1970
Russell: 3291 minutes to 0
Havlicek: 3174 minutes to 3369
Siegfried: 2560 minutes to 2081
Howell: 2527 minutes to 2078
Sanders: 2184 minutes to 1616
Jones: 1920 minutes to 0
Nelson: 1773 minutes to 2224
Bryant: 1380 minutes to 1617
Barnes: 595 minutes to 1049
Johnson: 163 minutes to 898
Chaney: 209 minutes to 839 minutes
Finkel: 0 minutes to 1866
White (rookie): 0 minutes to 1328

Bulls: 1998 -> 1999
Jordan: 80% minutes played to 0%
Rodman: 72% minutes played to 0%
Harper: 58% minutes played to 46%
Kukoc: 56% minutes played to 69%
Longley: 43% minutes played to 0%
Pippen: 42% minutes played to 0%
Brown: 29% minutes played to 48%
Kerr: 28% minutes played to 0%
Burrell: 28% minutes played to 0%
Wennington: 12% minutes played to 19%
Simpkins: 6% minutes played to 60%
LaRue: 3% minutes played to 30%
MBryant: 0% minutes played to 50%
BBarry: 0% minutes played to 49%
Dávid (rookie): 0% minutes played to 37%

Weird that we are playing around with this when Jordan was kind enough to give us large “without you” samples independent of five of the other top players on the team leaving…

[Djoker]

With a lot of players, especially in the modern NBA, sitting out second nights of back-to-backs and resting the last few games of the regular season, the low teams records in off samples can be misleading at times. And sometimes the team tanks when their superstar is out the entire season to get higher draft picks.

For example with Curry's off samples... there is important context that a lot of those missed games are from the 2019-20 season. Which not only Curry missed but also Klay. The Warriors pretty much tanked that year. Not saying Curry doesn't have great ON-OFF impact because he does but context is huge here.

[DraymondGold]

I was on the "Is Steph Curry Still Unappreciated" Thread and posted this:

As incredibly hyped as he's been, I still think he's perhaps still underestimated (in an ATG/historical sense):

All stats include the playoffs:

--Warriors since 2009-2010 without Curry playing: 76-134 (.362)
--Warriors since 2009-2010 with Curry playing: 643-328 (.662)

I challenge anyone to find a player with a really long career that has a bigger differential.

I was strongly encouraged to start a discussion just about this topic. I think it bears discussion.

So, who are the (other) most impactful players of all time, besides Curry, in terms of team winning percentage on/off or "WOWY"?

Huh, cool stat ty! I did a similar WOWY estimation for Curry in the Greatest Peaks project (https://forums.realgm.com/boards/viewtopic.php?p=100487575#p100487575)

It turns out there's different types of WOWY stats, just like there's different types of plus minus stats. It looks like you looked at how the record changes with and without a player -- but you could also look at how a team's point differential changes in games when a player plays vs when they don't. This is sort of like on-off stats, but here we're using margin of victory during a player's Games Played for the "on" sample and margin of victory during Games Injured/Rested as the "off" sample (rather than possessions played for the "on" sample and possessions rested for the "off" sample). In the link I posted above, I did an estimate (with a bit more uncertainty) for how Curry ranks all time, and he ends up 1st all time in this version of WOWY

The biggest issue is how to handle 2020; since Klay was also out for that year, the Warriors performed worse than if just Steph was out. So the two (lazy) options are to include 2020 but try to account for the fact that Klay was missing, or use the other non-2020 years. (or you could do a full regression to get the actual number, but that's a pain, so I'll just wait for Thinking Basketball to post it eventually).

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see. The biggest trends I've noticed are that WOWY tends to rank playmakers very highly (over play finishers / iso scorers), as well as defensive anchors (over perimeter defenders). My guess is because playmakers and defensive anchors are harder to replace in a night-to-night gameplan, they end up being missed more in the games they miss.

As others have pointed out 10 year samples can be incredibly noisy even if the amount of games is decent cause the "control" in the study (the teammates and rival quality) can wildly change fron a missed game one season to a missed game 8 years later.

Seems like focusing on single seasons with a big missed game sample (10+ at least imo) or full seasons "before and after" a player may be more useful

Example russel leaving the championship celtics in 70 (past his prime to boot) and the team falling to 30 wins. A drop off comparable to the 99 bulls losing pippen, jordan, rodman and phil jackson

Hiya falcolombardi! I have some agreements, and some disagreements. Let's go for where I disagree first:

The pushback: You suggest that long-term WOWY studies are too noisy, because it's too difficult to control for the teammate and opponent quality. To be blunt: I definitely disagree. We've been controlling for teammate and opponent quality for two decades now... otherwise getting from on/off to Adjusted Plus Minus or RAPM would be impossible. Controlling for teammate and opponent quality is the very thing we do when taking on/off data and calculating APM or RAPM.

It turns out you can do the same treatment with WOWY data. For example, you might use per-game SRS instead of MoV to account for opponent quality, and you might use a regression to find WOWY for every player on your team to account for teammate quality in games when multiple teammates sit out. These are the kind of corrections that Thinking Basketball actually applies when calculating 10-year Prime WOWYR. So to reiterate: WOWYR does indeed correct for opponent and teammate quality, just like RAPM does.

The agreement: That said, these kinds of corrections for teammate/opponent quality are difficult for people who just want to do a back-of-the-napkin approximation without using a bunch of code to solve a massive regression problem. As I openly admitted in my Curry WOWY estimation (see the link in my last post), I didn't do a careful consideration of varying opponent/teammate quality specifically because the problem gets a lot harder, which leads to the greater uncertainty I mentioned.

When people want a quicker alternative to the full teammate/opponent correction, I definitely agree that it can be good to look at scenarios where a player misses a lot of time (e.g. some late 60s years for Jerry West), or when a player has just been drafted/traded/retired.

But: my one caution for this latter method (using single-season samples when a player is drafted/traded/retires) is to make sure you're doing an apples-to-apples comparison. Looking at when Russell retired after 69 vs when Jordan retired after 98 might tell you something about the WOWY value of 68 Russell and 98 Jordan, but that doesn't necessarily apply to 64 Russell vs 91 Jordan. The nice thing about 10-year samples is you're more likely to be comparing apples-to-apples (Russell's 10 year prime vs Jordan's 10 year prime).

And as people have mentioned, you want to be careful to consider context either way... for example, Stockton' only missed 22 games in his 10-year prime (and they only came at the start/end of his prime), so Stockton's off sample might be extra noisy/wonky.

Here's the all-time rank for 10-year prime, according to WOWY:
1. 2012-2022 (no 2020) Curry: +10.3 (with more uncertainty, since this is an approximation)
1. 2013-2022 (with 2020, but adjusting for Klay's absence): +10.2 (with more uncertainty, since this is an approximation)

2 tie. Prime Russell: +9.4
2 tie. Prime Robinson: +9.4
2 tie. Prime Nash: +9.4
5. Prime Magic: +8.3
Jordan's 8th all time, and LeBron's a bit lower after that. WOWY's definitely a noisy stat, but it's still interesting to see.

This is Ben’s (well, not strictly Ben’s, but he is the one who posted it) “Game-Level Adjusted Plus/Minus” (GPM) metric, not WOWY or even his own WOWYR (or its various iterations).
https://thinkingbasketball.net/tag/wowyr/

You're quite right! I used GPM over WOWYR because GPM's a bit more of a "pure"/simple statistic (at the possible cost of being a little noisier/less consistent), so it was closer to my ballpark estimation for Curry. I mentioned that the list is from GPM in the post I linked. Perhaps I could have used just WOWY instead of GPM or WOWYR (which is sort of parallel to using on/off instead of APM or RAPM), but I wasn't sure where to find a list of Prime WOWY.

Which leads me to my second question...

Purer “prime” WOWY is more like (Curry? >) Garnett > Lebron > Shaq > Nash > Bird > Hakeem > Magic > Robinson > Duncan > Kobe > Jordan. With the acknowledgment that the purer form is even more variable, it does still correlate reasonably well to longer RAPM and on/off studies, with Lebron, Garnett, and Shaq (and Curry) maintaining as the top ~ten-year prime performers since 1994/97.

Thanks for the Prime WOWY list! Do you mind sharing the link? The only link I found was for Prime WOWYR and Prime GPM.

...

I think this possibly shows a flaw with building your entire defense or offense around one person because the moment they go to the bench it all falls apart. I do wonder how often this is by design (not investing much in a back-up because you want to play your main man as much as possible anyway and you still need to invest elsewhere too) or because of there simply being a shortage of good playmakers and defensive anchors available as back-ups. I feel like on/off stats could really benefit from the context of whether someone is breaking the stat because the team is built around them and their back-ups are terrible or if they're destroying the on/off stats despite having capable back-ups.

Strong agree here! it would be fantastic to have some more consistent way of telling how much plus minus is from terrible/ill-fitting back-ups or not.
[not to retread the oft-repeated LBJ-MJ debate, but this ties into the discussion from that '5-year On/off Study data for LBJ/MJ' thread a few months back.... MJ has a better 'on' numbers than LBJ, but LBJ's teams have a massively worse 'off' numbers. This gives LBJ better overall on/off, but one could argue that it's just because LBJ's teams had worse-fitting backups than MJ. Anyway...]

I do think at least some of WOWY's preference for playmakers/defensive anchors is from a shortage of good backups, more than "by design" as you say. Imagine you're on a Steve Nash team and Nash is sitting for one night. Nash pretty clearly runs your entire offense when he's on the court. So what do you do when Nash sits?
Option A: Adjust your rotation. You can have your bench point guard (e.g. Marcus Banks) go from ~11 minutes per game to ~35+ minutes per game, hope he survives the minutes, and hope the rest of the starting lineup adjusts to having a different primary playmaker.... but this is pretty clearly a poor idea.
Option B: Keep your rotations fairly steady, but redesign the offensive scheme of the starting lineup to adjust for the missing Nash. Perhaps you have a combination of more playmaking from Barbossa and Diaw, along with more iso from Amar'e. This is also a pretty poor idea if Nash is missing for just one night, as it can be pretty hard to change an entire offensive scheme overnight.
Option C: attempt some combination of A and B, but end up playing terribly and lose the game.
Most teams probably end up with option C. This will increase Nash's "off" sample in WOWY, thus giving him better WOWY numbers. You can imagine a similar process happening for other offensive playmakers and defensive anchors/leaders. My guess is this is (at least partially) why WOWY likes these kinds of players so much.