The Multi-Year WOWY Database

[DraymondGold]

Ooh good spot!

"adjusting for expansion" for one guy but not another is pretty bad practice. Would help explain Kareem's uncharacteristically poor performance here.

Don’t think he made an “adjustment” but to make a note of it on one player, but then not on another player… Doesn’t look good. It certainly doesn’t scream objectivity.

ah, okay.

Dray is probably right expansion was more notable. But it is a factor for both.

Now that, OhayoKD, is a civil way to discuss with someone who said something you disagree with (or more accurately, who didn't say something you wish they had). I'll add the comment.

[homecourtloss]

Now this seems like multiple straw man arguments at once...

1) Expansion
The expansion in 70 was far bigger and produced far larger spread in SRS than the one in the late 90s. And 70 Kareem was doubly benefiting from the expansion -- the Bucks' very first first year in existence was their 'without-Kareem' sample (with plenty of contextual reasons for why their SRS was poor and likely to improve), then when they did improve they had plenty of poor teams to beat up on (and boost their SRS). You can look at the spread of SRS in sansterre's Top 100 list and see the difference between the early 70s and the late 90s.

If you are presenting a dataset, everything has to be equal, and the more things that you include that are not equal, the less objective the data set seems. Now you are talking about the level of expansion; perhaps you could have noted that in the initial post for why you were including that for Kareem, but not for Jordan. But overall, it doesn’t matter—if you note that there is an expansion for one player and then not for another player in the same data set, well, that doesn’t seem very objective.

2) Coasting
I made it pretty obvious in most post on the last page that I was marking the coasting adjustment for Jordan fans who would ask, and that I did *not* include the coasting adjustment when doing primes or career averages. And yet there's

It is hard to tell which numbers you used for your prime, peak, lists at the end because you simply present the alternate and adjusted and contextual, etc., numbers in the player profiles, and then you have your prime, peak numbers at the end and then who knows which numbers were used. This is why I asked for all the numbers without any of the alternate an adjustment numbers. Also, it doesn’t matter what certain fans asked for or not — if you are going to create a new AlterNet, or adjusted or whatever value for “coasting” for one player, but not for the other players, then it doesn’t seem very objective.

3) Arbitrary game sample.
If you'd like to do the work for a lower game threshold, be my guest.
In the future, perhaps some more productive posting for you might be
-Kindly asking to add that I add a comment about the late 90s Expansion if you think it's important. Plenty of people have argued the 70s expansion was more important factor for those teams, so it's not unreasonable to mention that first
-Kindly point out other years that are worth adding for coasting. 92 vs 93 Jordan was discussed pretty extensively in the Top 100 project, and I'm not aware of any other coasting arguments at all that project. So again it's not crazy to mark that some people believe coasting is a factor

I made it pretty clear that I'm willing to improve the database with helpful comments... so phrasing all these as if they're "gotcha's" seems to have missed the point entirely...

This seems like a very defensive comment. I did ask twice and instead of listing the numbers, I saw more justification for random adjustments and context and other things.

Umm... what?

The adjustment for 98 Jordan literally *explicitly says I subtract out Pippen and Rodman's WOWY*... did you miss that? Of course Pippen and Rodman had zero minutes in 1999. That's why I subtracted out their WOWY from Jordan's. You have multiple players that leave. So you look at the net drop, break up how much each other player contributed, then the remainder goes to the player you're looking at. Seems pretty simple really.

Yes, and this was done correctly, and I mentioned that, but the problem was was that this was adjustment was represented as exactly the same type of adjustment that you would make for LeBron’s 2010-2011 numbers.

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

The adjustment for Jordans number here makes sense because Pippen did not play a single minute for them, Rodman didn’t play a single minute for them, and Phil Jackson, perhaps the greatest coach ever, did not coach them anymore. But then you have listed here in the very same way some adjustments for over the hill Shaquille O’Neal (I don’t see any other adjustments for players of 2010 Shaquille O’Neal level — why is that?), and then Mo Williams and Anderson Varejao as if the situation is exactly the same. Why would you do this? You say that:

As for why I'm adjusting for "Andy V. and especially Mo Williams", when there players who are top in minute, say Top 3, who also happen to miss a lot of games, you want to adjust for them. Those two players are in the top of the Cavs rotation, and they missed plenty of games. So I adjust for them. Note that this something I checked and for every single player. You'l find comments throughout noting that I did the same thing for Kareem, the same thing for Jordan, the same thing for Duncan, the same thing for Curry....

They were just as bad with these players playing — they didn’t make any difference at all. The +15 or so SRS number is more indicative of the impact than the “adjusted alternate” number but yiu choose your alternate health adjustment number. Why would you use the alternate number in this case rather than the one that more accurately describes what happened? Mo Williams Andy V did this:

We have a 21 game sample with mostly same players that Taylor used and in that sample: 19 win pace

Andy V: -9.0 ON, +.5 ON/OFF, 8-23 in games played, 19 win pace, didn't do anything
Mo Williams: -13.9 ON, -4.4 ON/OFF, 9-28, 20 win pace (Cavs were better with Mo off court)
Mo Williams + Andy V.: 27 games played, -9.5, 6-21, 18 win pace;

they were worse together than the “before adjustment” value fur the Cavs.

Also, why was Shaquille O’Neal taken into account here? Did you do that for every other team for a player who played the minutes that Shaquille O’Neal did, and did nothing and retired after one more year? Are the other numbers adjusted for some player like 2010 Shaquille O’Neal? What about an adjustment for Anderson Varejao in 2015 when he missed 50+ games? There’s not even a mention of injured there. This is what I mean about consistency.

As for going through every player and seeing players who missed minutes, well, you just said you missed someone like 1997 Shaq and his minutes, so I don’t know.

[lessthanjake]

it is a much simpler assumption than what happens when you claim player a's team was loaded because it had a guy in the top 100 or player b's team was as good because of the top names on a roster.

It has not been presented as a certainty. You seem to be freaked out by the numbers but I'd say they're an improvement over "raw vibes"

It’s the output of numbers that make what you’re doing objectionable!

So if I said "jordan did not do as well as player a with similar or more help" it would be valid?

Per usual the criticism is aesthetic

The number is specific and offers something for you to scrutinize. The assumptions are laid out pretty plainly. I would think someone vigilantly looking for "motivated reasoning" would appreciate that...

Yes, the number is specific, and it is also completely meaningless and that’s what I keep pointing out. You’re welcome to make a more qualitative point about the success of Jordan's early teams, but trying to make a quantitative point in the way you keep doing is just plain silly. It’s essentially just a number taken completely out of thin air, and so there's no way to meaningfully discuss it beyond pointing out that your methodology is complete nonsense and therefore the number spit out has zero probative value.

[OhayoKD]

It’s the output of numbers that make what you’re doing objectionable!

So if I said "jordan did not do as well as player a with similar or more help" it would be valid?

Per usual the criticism is aesthetic

The number is specific and offers something for you to scrutinize. The assumptions are laid out pretty plainly. I would think someone vigilantly looking for "motivated reasoning" would appreciate that...

Yes, the number is specific, and it is also completely meaningless and that’s what I keep pointing out

To you.

Luckily we are not bound by what you consider useful. Not everyone goes by the vibes, you know

[lessthanjake]

So if I said "jordan did not do as well as player a with similar or more help" it would be valid?

Per usual the criticism is aesthetic

The number is specific and offers something for you to scrutinize. The assumptions are laid out pretty plainly. I would think someone vigilantly looking for "motivated reasoning" would appreciate that...

Yes, the number is specific, and it is also completely meaningless and that’s what I keep pointing out

To you.

Luckily we are not bound by what you consider useful. Not everyone goes by the vibes, you know

I don’t think there is anyone but you that finds it meaningful (except maybe ShaqAttac haha).

[AEnigma]

“Meaningful” is a strong word, but I find it about as interesting to consider as “take 1998 Pippen’s WOWY and 1997 Rodman’s WOWY and give the remainder to Jordan.” Which is to say: moderately interesting and worth articulating as an approach.

[ShaqAttac]

Yes, the number is specific, and it is also completely meaningless and that’s what I keep pointing out

To you.

Luckily we are not bound by what you consider useful. Not everyone goes by the vibes, you know

I don’t think there is anyone but you that finds it meaningful (except maybe ShaqAttac haha).

u really got no self awareness huh

[Bklynborn682]

yea these adjustments seem kinda sus

While I appreciate the OP’s effort, this is what I was getting out when I was discussing about the “adjusted” and “alternate” values with arbitrary players picked for “adjustments.” If you are going to do it, then you have to be consistent all the way around or present a list that doesn’t have any of the adjusted and alternate values. There are arbitrary decisions made and then there’s final numbers presented, which is misleading.

The Lakers were roughly +5 SRS (+/- .5) with Shaq in a large sample without doing the calculations, so saying Lakers were -.55 after adding Shaq isn’t an accurate depiction.

So to start off, this is a genuine mistake with Shaq. I didn't catch that he was injured in 97 and missed a big chunk of games. I mentioned that this was by hand and thus could be subject to mistakes (and to kindly inform me if so), so thanks to Bklynborn682 for catching this! I double checked every other person's except Shaq's, who I missed out of laziness... I guess that's what I get. Fixed now though!

You mentioned about Kareem:

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)

Note that there is a comment about “not correcting for expansion,” but no such comment on Jordan’s profile:

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]
-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

Here you have a bunch of health adjustments, context adjustments alternate values, including a subjective “coasting” argument (have to be consistent and apply “coasting” everywhere and not when convenient for whatever purposes), etc., so you’d think there would be something there for the 1995 Bulls with Jordan just for the record, even if it doesn’t meet the arbitrary (and convenient) 30 game sample.

Now this seems like multiple straw man arguments at once...

1) Expansion
The expansion in 70 was far bigger and produced far larger spread in SRS than the one in the late 90s. And 70 Kareem was doubly benefiting from the expansion -- the Bucks' very first first year in existence was their 'without-Kareem' sample (with plenty of contextual reasons for why their SRS was poor and likely to improve), then when they did improve they had plenty of poor teams to beat up on (and boost their SRS). You can look at the spread of SRS in sansterre's Top 100 list and see the difference between the early 70s and the late 90s.

2) Coasting
I made it pretty obvious in my post on the last page that I was marking the coasting adjustment for Jordan fans who would ask, and that I did *not* include the coasting adjustment when doing primes or career averages. And further actively did *not* encourage people taking the prime rankings too seriously.

3) Arbitrary game sample.
If you'd like to do the work for a lower game threshold, be my guest.

All 3 of these straw mans seem strange... because there's perfectly reasonable alternative ways you could have gone about things...In the future, perhaps some more productive posting for you might be
-Kindly asking to add that I add a comment about the late 90s Expansion if you think it's important. Plenty of people have argued the 70s expansion was more important factor for those teams, so it's not unreasonable to mention that first
-Kindly point out other years that are worth adding for coasting. 92 vs 93 Jordan was discussed pretty extensively in the Top 100 project, and I'm not aware of any other coasting arguments at all that project. So again it's not crazy to mark that some people believe coasting is a factor
-Offer to help if you think it's important to lower the game sample, rather than implicitly accusing someone of bias of making a "arbitrary and convenient" threshold minimum (as if I made this threshold specifically because I really wanted to buff up Jordan ), while offering to do no work to address the problem you see.

I made it pretty clear that I'm willing to improve the database with helpful comments... so phrasing all these as if they're "gotcha's" seems to have missed the point entirely...

Then you have the LeBron sample:

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

What’s going here in blue? Adjusted for washed Shaq? Also, Andy V. played in 2011; Mo Williams played in 2011. They even played together and they did absolutely nothing. If you’re going to adjust for players like Andy V. and especially Mo Williams here, then you have to go through this entire list finding equivalents otherwise it seems biased, i.e., why are you picking out these two for LeBron’s sample? 1999 Bulls had ZERO minutes from Pippen, 0 minutes from Rodman, zero from Kerr and perhaps the greatest coach ever was gone. You need some adjustments there especially since 1999 Pippen was still an impact force. How is this equivalent to actually getting 2000+ minutes from Andy V. and Mo?

Umm... what?

The adjustment for 98 Jordan literally *explicitly says I subtract out Pippen and Rodman's WOWY*... did you miss that? Of course Pippen and Rodman had zero minutes in 1999. That's why I subtracted out their WOWY from Jordan's. You have multiple players that leave. So you look at the net drop, break up how much each other player contributed, then the remainder goes to the player you're looking at. Seems pretty simple really.

As for why I'm adjusting for "Andy V. and especially Mo Williams", when there players who are top in minute, say Top 3, who also happen to miss a lot of games, you want to adjust for them. Those two players are in the top of the Cavs rotation, and they missed plenty of games. So I adjust for them. Note that this something I checked and for every single player. You'l find comments throughout noting that I did the same thing for Kareem, the same thing for Jordan, the same thing for Duncan, the same thing for Curry....

So acting like this is some special thing just to ding LeBron's value down (while skipping over comments and missing pretty blatant text throughout the rest of the post) just seems intentionally antagonistic. I'm willing to make improvements to the Database if you're constructive, I made it clear this is a work in progress and requires input from the community to check for things I missed, but this just seems bent on being argumentative while actively missing things that are already there that address a number of your concerns.

Not a big deal but I was wondering since you adjusted 97 Shaq did you adjust for magics retirement and his wowy?

[DraymondGold]

.

Hey lessthanjake! As a fellow basketball statistics fan, I thought you mind find this interesting. I'm definitely *not* wanting to turn this thread into another big Jordan argument. Moreso, I was interested in analyzing uncertainty in single-season WOWY samples, and I thought I would use 1986 Jordan as a starting place. It would be interesting to see some similar uncertainty analysis on other GOAT/All-time level players!

...

In 1986, Jordan broke his foot and missed a large sample of games. This provides one of his only WOWY samples in his prime. How does Jordan look?

Method 1: raw WOWY
1986 Jordan raw WOWY: -3.61 MoV with, -3.86 MoV without, +0.25 MoV change
1986 Jordan raw WOWY: -4.87 SRS with, -4.82 SRS without, -0.04 SRS change
*Note: this does not control for other changes in the Bulls lineup (Orlando Woolridge, 1st in Bulls minutes, also missed 12 games this season). Doing so would give Jordan a +2.0 SRS change (Method 2: WOWY, controlling for other lineup changes)
*Note this also does not control for home court advantage or diminishing returns on better teams, which more intelligent WOWY stats do. Doing so would give Jordan a +1.2 WOWY, with a +/- 0.8 uncertainty for the without sample, and a +/- 5.0 uncertainty for with sample.

But there’s another problem with the 1986 WOWY sample: Jordan’s minutes. Here’s Jordan’s MPG by season:
1985: 38.3
1986: 25.1
1987: 40.0
1988: 40.4

By linear interpolation, we’d expect 1986 to be just over ~39 MPG (really 39.15). Instead he played at 25.1 MPG.

Why? He was on serious minute restrictions. When he returned from injury, these were his minutes in each game, going from his first game lack to his last: 13, 14, 15, 16, 16, 19, 22, 23, 26, 28, 31, 31, 33, 37, and 29. There’s a very obvious trend, increasing his minutes each game until the final game when he rested more before the playoffs. He didn’t even start until the last 4 games.

The problem with a plain raw WOWY analysis of 1986 is that WOWY is blind to this context. If we don’t control for this, 39% of our ‘With’ sample for Jordan ends up having Jordan play *less than 50% of his expected minutes*. This isn’t really measuring the true WOWY change of healthy sophomore Jordan.

Moonbeam found he preferred a threshold of only using 18+ MPG players as the best threshold for whether a team was with or without a player. Thinking Basketball found longer term WOWY scores were most accurate setting a threshold at ~25 MPG. Now these minute thresholds are optimized over 5 and 10 year timeframes specifically, rather than 1 year timeframes. There’s a relationship between the minute threshold and your With/Without sample size, and the resulting uncertainty, so without testing it’s not clear what the optimal single-season threshold is. But it does suggest that low MPG games can decrease the accuracy of WOWY, and they do give a precedent to try setting a minute threshold, even if they don’t tell us what the ideal minute threshold is for a single-season sample.

For simplicity, let’s try taking their minute thresholds for this single season sample, acknowledging that these threshold may not be optimized for a single-season sample.

Method 3: WOWY, with a more intelligent 18 minute threshold
1986 Jordan WOWY: 0.54MoV with, -4.62 MoV without, +4.16 MoV change.
1986 Jordan WOWY: -0.39 SRS with, -5.81 SRS without, +6.20 SRS change
Note: we now have only a 13-game with sample, which would have a 95% uncertainty of over 6.2.

Method 4: WOWY, with a more intelligent 25 minute threshold
1986 Jordan WOWY: -1.22 MoV with, -4.12 MoV without, +2.9 MoV change
1986 Jordan WOWY: -2.96 SRS with, -5.06 SRS without, +2.10 SRS change
Note: we now have only a 9-game with sample, which would have a 95% uncertainty of over 7.7.

Jordan’s raw WOWY gives us a (noisy) signal for the difference between a team without Jordan and a team with Jordan playing 25.1 MPG. But if Jordan were truly healthy, we would expect him to be playing ~39 MPG, 55% more. Thus we could also try boosting Jordan’s ‘with’ signal by 55%, to estimate Jordan’s value if he were healthy and not minute-restricted.

Method 5: WOWY, curving Jordan's ‘with’ sample up to estimate the healthy 39 MPG value.
1986 Jordan WOWY: -2.33 MoV with, -3.86 MoV without, +1.53 MoV change
1986 Jordan WOWY: -3.14 SRS with, -4.82 SRS without, +1.68 SRS change

And again, these don’t control for other changes in the Bulls lineup (Orlando Woolridge, 1st in Bulls minutes, also missed 12 games this season), which would make an estimated 39 MPG Jordan look better.
Method 6: WOWY, with the same minute curving as Method 5, while also controlling for other lineup changes like in Method 2
1986 Jordan raw WOWY (controlling for Woolridge playing too): -1.10 SRS with, +2.6 SRS change

Rather than just boosting the ‘with’ sample by 55% to get from 25.1 MPG to 39 MPG, we could also try taking a linear fit for Minutes vs MoV or SRS. We could then compare the best-fit MoV and SRS when Jordan’s playing 0 MPG vs his expected 39 MPG.
Method 7: WOWY, estimating Jordan's 39 MPG value using a linear fit
1986 Jordan WOWY: -2.33 MoV with, -4.18 MoV without, +2.67 MoV change
1986 Jordan WOWY: -2.55 SRS with, -5.21 SRS without, +2.65 SRS change

So to summarize this survey of 7 possible methods:
Jordan’s raw WOWY change: +0.25 MoV, -0.04 SRS
Jordan’s WOWY change using a potentially more intelligent sample: +2.0 SRS (controlling for Woolridge)
Jordan’s WOWY change using a potentially more intelligent sample: +4.16 MoV, +6.20 SRS (18 MPG threshold)
Jordan’s WOWY change using a potentially more intelligent sample: +2.99 MoV, +2.19 SRS (25 MPG threshold)
Jordan’s WOWY change using a potentially more intelligent sample: +1.53 MoV, +1.68 SRS (estimating 39 MPG Jordan ‘with’ sample)
Jordan’s WOWY change using a potentially more intelligent sample: +2.60 SRS (estimating 39 MPG Jordan ‘with’ sample, controlling for Woolridge)
Jordan’s WOWY change using a potentially more intelligent sample: +2.67 MoV, +2.65 SRS (estimating 39 MPG Jordan using linear fit)

This is a clearly highly noisy single sample. Small changes in the methodology produce vast changes in the signal, ranging from a -0.04 WOWY to a +6.2 WOWY. Which makes sense if we’re dealing with a single-year sample that has a 95% uncertainty of over +/- 5.0, minimum. Heck, even just removing his first game back (when he played 13 minutes total) from his 'With' sample would have him go from a -0.04 SRS change (see method 1) to a +0.91 SRS change.

All that to say, it’s not clear which method is best. It would likely require testing across a much larger sample of players to find the most accurate methodology for measuring single-season WOWY, when applying some sort of minute threshold or minute weighting. But regardless of which method is best, there is a universal trend here. Any attempts to use more intelligent samples, such as controlling for other significant changes in the Bulls lineup or controlling for Jordan’s significantly lower minutes will produce a *significantly better* WOWY sample for sophomore Jordan than just looking at the raw WOWY alone.

I see this as a reminder of the uncertainty in using just one single-season WOWY sample to analyze a player. It's important to remember that the uncertainties can be quite large, and contextual analysis such as correcting for biases in minutes or for other major lineup changes may shift the WOWY rating up or down. This isn't to invalidate WOWY or anything. I do believe there's value in WOWY data: there is a signal in the noise, and it's a very enticing signal, as it can measure most of a player's wholistic value. But it's a good reminder of the range of uncertainties and the importance of context. Thoughts?

[Djoker]

Great work DraymondGold.

I would like to see David Robinson if you ever get to doing a few more players.

[DraymondGold]

~Ballparking Noise in 2-year WOWY samples~

Traditional WOWY samples (team performance with and without someone playing from a given season) is quite noisy due to small samples. Two-year WOWY samples, e.g. when a player joins a new team or gets traded or retires, can increase the sample size. But is this data significantly less noisy? We can estimate this by comparing overlapping two-year WOWY samples.

Let’s look at all the times when an all-time player changed teams in the off-season. This will give us two multi-year WOWY samples (one WOWY sample from the team the player left, and one from the team they joined). We can then compare these samples. For example: In 1968–69, Wilt Chamberlain left the 76ers and joined the Lakers. This gives us two WOWY samples to compare, one from when he left the 76ers (7.96 SRS with in 68, 4.79 SRS without in 69, total change: +3.17) and one when he joined the Lakers (3.84 SRS with in 69, 4.99 SRS without in 68, total change: -1.15). Notice how much the score changes, despite using overlapping two-year WOWY samples.

In this database, there were 14 times when a Top 15 player left a team and joined a new one in adjacent years (so 14 times *2 teams involved = 28 two-year WOWY samples). How does this look across the full database?

Spoiler:

This data is taken directly from the OP. See there for better formatting / more details.
Year / Player / WOWY when leaving a team / (Leaving Adjusted for major roster changes) / WOWY when joining a team / (Joining Adjusted for major roster changes) / Difference between two WOWY scores (Difference between adjusted scores)
1965–66 Wilt Chamberlain 2.29 (8.7) 0.78 (3.49) -1.51 (-5.21)
1968–69 Wilt Chamberlain 3.17 -1.15 -4.32
1970–71 Oscar Robertson 0.41 ` 7.66 7.25
1975–76 Kareem Abdul-Jabbar 1.8 4.12 2.32
2001–02 Hakeem Olajuwon 7.02 -2.39 -9.41
1996–97 Shaquille O’Neal 5.47 1.64 (2.22) -3.83 (-3.25)
2004–05 Shaquille O’Neal 6.67 5.89 -0.78
2008 Shaquille O’Neal 9.26 -2.62 -11.88
2009–10 Shaquille O’Neal 1.7 -2.51 -4.21
2010–11 Shaquille O’Neal 15.05 (7.63) 1.46 -13.59 (-6.17)
2007–08 Kevin Garnett 3.1 13 (9.3) 9.9 (6.2)
2010–11 LeBron James 15.05 (10.94) 4.77 -10.28 -6.17
2014–15 LeBron James 7.07 7.94 0.87
2018–19 LeBron James 9.98 0.11 (1.09) -9.87 (-8.89)

-Average score: +4.53
-Average magnitude difference in overlapping two-year WOWY scores: 6.53 (144%)

So in this set of overlapping two-year WOWY samples, the variability is over 100%! Those are massive uncertainty bars! This suggests two-year WOWY samples are still *extremely* noisy, despite the increased sample size. Where does this come from? Players may not change much in adjacent years (i.e. 1969 Wilt is likely not 144% worse than 1968 Wilt), but they do change a little. Moreover, the rest of the players are likely to change as well. Coaching and league situations can change in adjacent years. Fit and role can impact WOWY. But of course, all of this is not enough to explain such large uncertainty bars — some of this has to be statistical noise.

Can we improve the noise by controlling for some other major roster changes during this timespan? (see OP for details)
-Average score, adjusting for some other major roster changes: +4.14
-Average magnitude difference in overlapping two-year WOWY scores: 5.79 (140%).
So manually adjusting for other major roster changes helps, but the variability is still >100%.

One thing you may notice — which could absolutely still be a small sample size issue, given how wide the uncertainty bars are (uncertainty range is >100% of the sample value, and we only have 14 pairs of overlapping 2-year WOWY to compare) — is that players tend to have a better WOWY sample when leaving a team vs when joining a team. This is not guaranteed, some players look better joining, but there is a trend in the current data.

-Average score when leaving – average score when joining a team: +3.53 WOWY
-Average score when leaving – average score when joining a team (adjusted for other major roster changes): +2.66 WOWY

10/14 samples were better when leaving a team one year vs joining a team the next year. These samples were all taken in adjacent years. I’d say the following 9/14 players are clearly in their primes in both years (65-66 Wilt, 68-69 Wilt, 70-71 Oscar, 75-76 Kareem, 96-97 Shaq, 07-08 Garnett, 10-11 LeBron, 14-15 LeBron, 18-19 LeBron). Although only 4/14 samples are pre-peak (using RealGM Peak project for peak years), this trend remains when we look only at prime samples or pre-peak samples.

This pattern of WOWY leaving a team producing better results than WOWY joining a team also appears when looking at Rookie Year vs Retirement WOWY, to a lesser extent. The average Retirement WOWY in this database was +4.74, while the average rookie WOWY (discounting LeBron and Curry since they haven’t retired yet) was +4.25. And this is with ~9/13 players being closer to their prime in their rookie year (Wilt, Oscar, Kareem, Bird, Hakeem, Shaq, Duncan, Garnett, Kobe) vs their retirement year (Russell, West, Magic, Jordan), at least in my estimation.

We continue to see this trend of 1) noise of >100%, and 2) usually better WOWY when a player leaves vs when they join a team, if we expand the sample to include the overlapping multi-year WOWY data from the rest of the Top 30 players. I’ll be using the latest RealGM Top 100 project for the top 30. For ease, I’m just calculating the WOWY samples that are in the players’ primes. This gives us 11 pairs of overlapping multi-year WOWY samples from Chris Paul, Kevin Durant, Julius Erving, Moses Malone, Steve Nash, Charles Barkley, and James Harden.

Spoiler:

Prime Chris Paul
2011–12: +1.28 SRS with, -3.11 SRS without, +4.39 (traded leaving)
2011–12: +3.05 SRS with, -3.13 SRS without, +6.18 (traded joining)
2017–18: +7.52 MoV with, -0.04 MoV without, +7.48 (traded leaving)
2017–18: +10.16 MoV with, +5.77 MoV without, +4.39 (traded joining)
leaving average: 5.94
joining average: 5.29 (-0.65 worse)
average magnitude difference: +2.35

Prime Kevin Durant:
2016–17: +8.65 MoV with, +0.76 MoV without, +7.89 (traded leaving)
2016–17: +12.35 MoV with, +10.76 MoV without, +1.59 (traded joining)
[no Brooklyn sample because of the year gap and no sufficient Brooklyn-Suns sample]
leaving average: +7.89
joining average: +1.59 (-6.3 worse)
average magnitude difference: +6.3

Prime Julius Erving
1973–74: -0.38 SRS with, -4.38 SRS without, +4.0 (traded leaving)
1973–74: +4.80 SRS with, -5.80 SRS without, +10.6 (traded joining)
1976–77: +2.56 SRS with, -6.54 SRS without, +9.1 (traded leaving; NBA merger)
1976–77: +3.78 SRS with, 0.33 SRS without, +3.45 (traded joining; NBA merger)
leaving average: +6.55
joining average: +7.03 (+0.48 better)
average magnitude difference: +6.13

Prime Moses Malone:
1982–83: -0.39 SRS with, -11.12 SRS without, +10.72 (traded leaving)
1982–83: 7.53 SRS with, 5.74 SRS without, +1.79 (traded joining)
leaving average: +4.14
joining average: -0.13 (-4.27 worse)
average magnitude difference: +4.27

Prime Steve Nash:
2004–05: 4.86 SRS with, 5.86 SRS without, -1.0 (traded leaving)
2004–05: 7.08 SRS with, -2.94 SRS without, +10.02 (traded joining)
leaving average: -1.0
joining average: +10.02 (+11.07 better)
average magnitude difference: +11.07

Prime Charles Barkley:
1992–93: -1.34 SRS with, -5.25 SRS without, +3.91 (traded leaving)
1992–93: 6.27 SRS with, 5.68 SRS without, +0.59
leaving average: +3.91
joining average: +0.59 (-3.32 worse)
average magnitude difference: +3.32

Prime James Harden:
2012–13: +6.44 SRS with, 9.14 SRS without, -2.70 (traded leaving)
2012–13: +3.69 SRS with, 0.57 SRS without, +3.12 (traded joining)
2020–21: 2.88 MoV with, -8.46 MoV without, +11.34 (traded leaving)
2020–21: +5.42 MoV with, 0.81 MoV without, +4.61 (traded joining)
2021–22: 3.29 MoV with, 0.34 MoV without, +2.95 (traded leaving)
2021–22: +5.81 MoV with, 3.71 MoV without, +2.1 (traded joining)
leaving average: +3.86
joining average: +3.28 (-0.58 worse)
average magnitude difference: +4.47

Across this full new sample:
-Average score: +4.84
-Average magnitude difference in overlapping two-year WOWY scores: 5.03 (104%)
-Average score when leaving – average score when joining a team: +0.45 WOWY
-7/11 samples are better when the player is leaving vs when they’re joining a new team

Adding this new sample in, we now have 25 pairs of overlapping pairs of multi-year WOWY samples from the Top 30 players. Adding in this new data, we have
-Average magnitude difference in overlapping two-year WOWY scores: 124%
-Average score when leaving – average score when joining a team: +2.17
So adding more players, we have variability of greater than 100%. And there still seems to be a bias: 17/25 WOWY scores when leaving a team produce better scores than when joining a team, although this pattern is not guaranteed. I’m certainly open to arguments that contextual factors are causing the apparent bias for ‘leaving a team’ WOWY samples. Perhaps the database is still small, perhaps players tend to leave teams only in specific situations that boost their WOWY score (e.g. the team tanks the following year), perhaps Top 30 players tend to leave teams more frequently after their peak and thus look worse in the following year, perhaps other contextual changes to teammate and coaching create this effect. But if any of these arguments are true, this gives further credence to the idea that any single two-year WOWY sample is very noisy, really too noisy to function as any sort of precise measure of a player’s true value.

Thus, it seems like even though multi-year WOWY data does increase the sample size, a variety of contextual changes still make it extremely noisy, with uncertainty bars of over 100%. To me, this suggests that it’s still best to use much larger WOWY samples over individual noisy samples (e.g. 10 year prime WOWY samples, with plenty of With and Without games), along with other analysis. And if possible, it’s better to adjust for teammates than to use raw WOWY.

[Djoker]

^Good post!

I think players leaving a team produce larger WOWY differences because often a team losing a superstar will tank for better draft picks. Once a team isn't a contender, there is little incentive to muddle in mid-tier mediocrity.

Player role and fit on the roster make a massive different in their impact as well. I'm not at all surprised by the big differences.

I see WOWY as just a simple indicator. For instance saying that a +8 WOWY is much better than a +7 WOWY or even a +5 WOWY is stretching it quite a bit. You can see which players are impactful and which aren't but how impactful is tough to say based solely on WOWY. It requires analysis of circumstances that is always subjective to an extent.