Winshares,winshares/48 and on/off metrics

[migya]

I think if you were to just go by like a summation of career value I kind of get that, but I also Think it just is a tad bit absurd thinking what’s gone on over the past 7-8 years, even if it’s cheesy or narrative driven I find it hard to compare someone thats probably had the most success in the nba since he hit his stride vs a guy who gets memed for the past few years lol

I think cp3 has been a victim of unfortunate circumstance to an extent for sure though

That is actually like someone in 2013 saying that Dwight had surpassed Garnett because of the past five years. There was much before that.

Curry literally started an offensive revolution and won 4 rings and been to 6 finals in the last 7 years, comparing that to dwight is almost as ridiculous as saying Curry is just a shooter.

Basketball is a team sport and Curry has has a good one for a while, Dwight not so much. Look at impact.

[OhayoKD]

You may have seen this before, but for posterity, this study builds off ben's base(or at least claims to):
https://dunksandthrees.com/blog/metric-comparison

so before i go any further, am i reading this table correctly?




they put in the metric of choice, use it to predict, and the RMSE represents the average error on the team net rating?

so like they use EPM, guess +2.48 net rating and then on average, since the RMSE is 2.48 (let's just assume RMSE and MAE are close enough for government work here), the team ends up being either a 0.0 (41 wins) or a +4.96 (54 wins)? so if they plug in WS48 and guess +2.48, it could be -0.37 (40 wins) or 5.35 (55 wins)?

Seems about right, yeah. Though maybe someone whose good at this stuff like unibro can check we're interpreting this properly

[SNPA]

Winshares is consistent among the stars. Chamberlain, Kareem, Robinson, Shaq, Malone, Barkley are some examples. I don't understand why, for example, it's considerably lower for Garnett while his on/off is high.

Noise. That is the term stat people like to use to explain away a data point they don’t like because it goes against their narrative. It’s not a sign the stat is junk, no not at all, couldn’t be. It must just be noise because the other data points of that stat fit with their narrative.

This has been today’s lesson on how stats are used subjectively to form an argument and are no more objective than the greatly mocked and horribly offensive “eye test” (the artist formerly known as watching the actual game of basketball).

[trex_8063]

Win Shares is fairly junky and archaic. One thing that's very obvious about Win Shares is that WS overweights Rebounding specifically Defensive Rebounding.



See: WS & WS/48 of '23 Nets Kyrie Irving compared to Nic Claxton. WS & WS/48 of '23 Cavs Darius Garland compared to Jarrett Allen. WS & WS/48 of '23 Celtics Jaylen Brown compared to Al Horford. WS & WS/48 of '23 Clippers Paul George compared to Ivica Zubac. The pattern is very clear.

I don't know that I agree WS over-values defensive rebounds (at least not to any noteworthy degree). I think the "pattern" you're seeing has more to do with WS placing heavy value on shooting efficiency (combined with a relatively [compared to PER] small consideration of scoring volume)......

Darius Garland's TS% is 57.8%.....Jarrent Allen's is 65.9%.
Kyrie Irving TS% is 60.5%.......Nic Claxton's is 71.2%.


Take a look at '21 Joe Ingles as an example......
In terms of rate [per 100 possessions] of all the box stats he was notably above average in assists/100 (league avg about 4.9, he was at 8.4), but otherwise: below average in pts/100, stl/100, blk/100, and notably below avg in reb/100 (league avg about 9.0, while Joe grabbed just 6.4 (specifically defensive rebs, league avg ~7.0/100, Joe getting just 5.7), while also commiting MORE tov/100 than league avg.

And yet his WS/48 was .180.

Granted this was for a team with a +9.25 MOV, but it still doesn't seem like that could possibly account for a .180 WS/48 when he's [collectively] a marginally below league avg player in terms of overall box production (and in particular: notably below league avg in the thing WS supposedly over-values more than anything [drebs]).

So what consideration is missing?
The shooting efficiency: he had 67.2% TS (+10% to league avg).


I think the "pattern" you believe you're seeing is simply because the guys who tend to have the highest TS% [low-moderate volume rim-runners] also tend to get lots of rebounds (because they're centers).

It's hard to say because Claxton has notably better rate of blocked shots, but there may even be a pinch of evidence in the examples you cited:
Jarrett Allen averages 15.4 reb/100 [10.9 DReb/100].....Claxton 14.7 [10.4]; and Allen's doing it for a team with a notably better MOV (+5.19 vs +3.09). Yet his WS/48 is slightly less (.196 vs Claxton's .202).

Again, hard to say because of the difference in blocks, but this could also be a suggestion of the value WS places on shooting efficiency (Claxton is +5.3% relative to Allen).

[f4p]

Win Shares is fairly junky and archaic. One thing that's very obvious about Win Shares is that WS overweights Rebounding specifically Defensive Rebounding.



See: WS & WS/48 of '23 Nets Kyrie Irving compared to Nic Claxton. WS & WS/48 of '23 Cavs Darius Garland compared to Jarrett Allen. WS & WS/48 of '23 Celtics Jaylen Brown compared to Al Horford. WS & WS/48 of '23 Clippers Paul George compared to Ivica Zubac. The pattern is very clear.

I don't know that I agree WS over-values defensive rebounds (at least not to any noteworthy degree). I think the "pattern" you're seeing has more to do with WS placing heavy value on shooting efficiency (combined with a relatively [compared to PER] small consideration of scoring volume)......

Darius Garland's TS% is 57.8%.....Jarrent Allen's is 65.9%.
Kyrie Irving TS% is 60.5%.......Nic Claxton's is 71.2%.


Take a look at '21 Joe Ingles as an example......
In terms of rate [per 100 possessions] of all the box stats he was notably above average in assists/100 (league avg about 4.9, he was at 8.4), but otherwise: below average in pts/100, stl/100, blk/100, and notably below avg in reb/100 (league avg about 9.0, while Joe grabbed just 6.4 (specifically defensive rebs, league avg ~7.0/100, Joe getting just 5.7), while also commiting MORE tov/100 than league avg.

And yet his WS/48 was .180.

Granted this was for a team with a +9.25 MOV, but it still doesn't seem like that could possibly account for a .180 WS/48 when he's [collectively] a marginally below league avg player in terms of overall box production (and in particular: notably below league avg in the thing WS supposedly over-values more than anything [drebs]).

So what consideration is missing?
The shooting efficiency: he had 67.2% TS (+10% to league avg).


I think the "pattern" you believe you're seeing is simply because the guys who tend to have the highest TS% [low-moderate volume rim-runners] also tend to get lots of rebounds (because they're centers).

It's hard to say because Claxton has notably better rate of blocked shots, but there may even be a pinch of evidence in the examples you cited:
Jarrett Allen averages 15.4 reb/100 [10.9 DReb/100].....Claxton 14.7 [10.4]; and Allen's doing it for a team with a notably better MOV (+5.19 vs +3.09). Yet his WS/48 is slightly less (.196 vs Claxton's .202).

Again, hard to say because of the difference in blocks, but this could also be a suggestion of the value WS places on shooting efficiency (Claxton is +5.3% relative to Allen).

WS loves efficiency, as you point out. see reggie miller having the same WS48 as hakeem, with basically 80% of it from offensive win shares, despite doing almost nothing else in volume that shows up in the box score and not even getting many DWS (see next point). it also loves when your teammates are better than you at defense because you siphon some of their defensive value, which almost certainly helps someone like ingles who was playing with gobert.

[migya]

So the consensus is the on/off metric is the best? Play by play looks erratic with much difference between some players' on court value to their on/off value. Seems rather arbitrary.

[70sFan]

So the consensus is the on/off metric is the best? Play by play looks erratic with much difference between some players' on court value to their on/off value. Seems rather arbitrary.

The consensus is that there is not a single stat that could tell you how good a player is/was. You need to do a lot more work than just to check some numbers.

[ty 4191]

So the consensus is the on/off metric is the best? Play by play looks erratic with much difference between some players' on court value to their on/off value. Seems rather arbitrary.

The consensus is that there is not a single stat that could tell you how good a player is/was. You need to do a lot more work than just to check some numbers.

Can you explain your own process, is some detail, 70sFan? Thank you!

[Owly]

Which players have the highest career on/off in the playoffs, all time?

Going off ben's videos(note it doesn't include non jordan pre-97 players), going by 3 year stretches you have Duncan/Drob topping, 16-18 Lebron second(among the players he shows), and then curry, shaq and jordan tied for 3rd iirc(shaq actually shows up at 2nd but then unibro and 70's found ben made a statistical error.

To clarify further, per 100 possessions.

A few notables (don't know how to sort this on BBRef):

-KG: +14.5 (career) Prime (97'-08'): +13.3
-LeBron: +10.8 (career) Prime (08'-18'): +13.1
-Duncan: +7.5 (career) Prime (98'-08'): +9.0
-Curry: +11.7 (career) Prime (12'-19'): +14.6

Anyone have on/off per 100 poss organized in a database for pre 96'-97' players?

Thanks!

Do you mean post 96-97- i.e. an organization of the data already out there?
Or are you asking for a database of data that isn't out there and for the most part probably doesn't exist? Suspect for the most part the data isn't there and to the extent it is, it's staying behind closed doors but maybe one can find snippets.

94-96 +/- (RS) is available on here year by year and from that they couldget a very good approximation of per 100 on-off but without the source pbp it's based on the assumption of pace being constant for on and off periods.

also there's
https://squared2020.com/2022/07/31/some-michael-jordan-plus-minus-numbers/

https://squared2020.com/2022/07/22/some-magic-johnson-plus-minus-numbers/

Maybe if you asked nicely you could inquire with Squared about particular stuff though I’m aware from a post I recall they may have other stuff going on and regardless aren’t obligated to share their work.


To the original question I’m not aware of it being sorted somewhere though it may just be me or it may be behind a paywall or whatever. However I looked up some big players I thought had / might have high on-off, some who do well in impact generally, plus some playoff production elevators plus some randos - This is by no means systematic, just ad hoc some quite good ones, no minutes qualifier:

Caveat: As ever noisy measure made noisier by playoffs uneven opponents (man and team level), uneven career playoff minute distribution, smaller (in some cases listed here tiny) samples

(yes formatting sucks, if anyone wants to copy and tidy feel free)
Player, "(inc)" if not full playoff career, on-off games in, minutes on
Blaylock (inc) 26.3 23 952
Brevin Knight 20.8 14 250
Robinson (inc) 18.9 70 2137
Baron Davis 15.8 50 1851
Garnett 14.5 143 5283
Bo Outlaw 14.4 22 416
Dr Green 13.7 145 5080
K Malone (inc) 13.6 96 3799
Shaq (inc) 11.7 180 6708
Lowry 11.1 107 3692
James 11 266 11035
Hornacek (inc) 11 61 1941
Divac (inc) 10.3 70 2131
Ginobili 10.2 218 6075
Br Russell (inc) 9.3 79 2573
Conley 9.2 73 2608
Hibbert 9.2 54 1708
Maxiell 8.7 35 581
Kidd 8.4 158 6088
Tracy Murray (inc) 8.4 5 92
P Gasol 7.9 136 4825
Stockton (inc) 7.9 75 2513
Bryant 7.6 220 8641
Duncan 7.5 251 9370
Dunleavy Jr 7.4 32 768
Horry 7 179 4522
Tim Thomas 6 55 1545
Jerome James 6 17 369
Devean George 5.9 86 1479
Paul 5.6 142 5192
Mutombo (inc) 5.5 86 2522
Nash 4.7 120 4289

late addition (and player lost in original post and then missed in do over and added at the last minute)
Marion 9.6 109 3837
Curry 11.5 134 4999
edit: New entry
Paul George 12 108 4222

[70sFan]

So the consensus is the on/off metric is the best? Play by play looks erratic with much difference between some players' on court value to their on/off value. Seems rather arbitrary.

The consensus is that there is not a single stat that could tell you how good a player is/was. You need to do a lot more work than just to check some numbers.

Can you explain your own process, is some detail, 70sFan? Thank you!

Not now, I don't have enough free time. I think it's nothing extraoridinary - watching a lot of games, looking at various on/off data and analyzing roster construction.

[MyUniBroDavis]

That is actually like someone in 2013 saying that Dwight had surpassed Garnett because of the past five years. There was much before that.

Curry literally started an offensive revolution and won 4 rings and been to 6 finals in the last 7 years, comparing that to dwight is almost as ridiculous as saying Curry is just a shooter.

Basketball is a team sport and Curry has has a good one for a while, Dwight not so much. Look at impact.

Saying to look at impact whilst you’re exposing why you don’t like impact data is a choice

[migya]

Curry literally started an offensive revolution and won 4 rings and been to 6 finals in the last 7 years, comparing that to dwight is almost as ridiculous as saying Curry is just a shooter.

Basketball is a team sport and Curry has has a good one for a while, Dwight not so much. Look at impact.

Saying to look at impact whilst you’re exposing why you don’t like impact data is a choice

Impact is overall performance, particularly with roster talent. Dwight had little around him, Curry had alot and Dwight was the best on his team at both ends and has some winning success.

[MyUniBroDavis]

Basketball is a team sport and Curry has has a good one for a while, Dwight not so much. Look at impact.

Saying to look at impact whilst you’re exposing why you don’t like impact data is a choice

Impact is liberal performance, particularly with roster talent. Dwight had little around him, Curry had alot and Dwight was the best on his team at both ends and has some winning success.

What the hell? Are you trying to argue Dwight over Curry now?

[OhayoKD]

Saying to look at impact whilst you’re exposing why you don’t like impact data is a choice

Impact is liberal performance, particularly with roster talent. Dwight had little around him, Curry had alot and Dwight was the best on his team at both ends and has some winning success.

What the hell? Are you trying to argue Dwight over Curry now?

what's "liberal performance"?

[migya]

Impact is liberal performance, particularly with roster talent. Dwight had little around him, Curry had alot and Dwight was the best on his team at both ends and has some winning success.

What the hell? Are you trying to argue Dwight over Curry now?

what's "liberal performance"?

Nothing to do with his of court tendencies.

Fixed it.

[OhayoKD]

What the hell? Are you trying to argue Dwight over Curry now?

what's "liberal performance"?

Nothing to do with his of court tendencies.

Fixed it.

gotcha

[OhayoKD]

Winshares is consistent among the stars. Chamberlain, Kareem, Robinson, Shaq, Malone, Barkley are some examples. I don't understand why, for example, it's considerably lower for Garnett while his on/off is high.

Noise. That is the term stat people like to use to explain away a data point they don’t like because it goes against their narrative. It’s not a sign the stat is junk, no not at all, couldn’t be. It must just be noise because the other data points of that stat fit with their narrative.

This has been today’s lesson on how stats are used subjectively to form an argument and are no more objective than the greatly mocked and horribly offensive “eye test” (the artist formerly known as watching the actual game of basketball).

who are you talking to?

[SNPA]

Winshares is consistent among the stars. Chamberlain, Kareem, Robinson, Shaq, Malone, Barkley are some examples. I don't understand why, for example, it's considerably lower for Garnett while his on/off is high.

Noise. That is the term stat people like to use to explain away a data point they don’t like because it goes against their narrative. It’s not a sign the stat is junk, no not at all, couldn’t be. It must just be noise because the other data points of that stat fit with their narrative.

This has been today’s lesson on how stats are used subjectively to form an argument and are no more objective than the greatly mocked and horribly offensive “eye test” (the artist formerly known as watching the actual game of basketball).

who are you talking to?

The world.

Happy New Years.

[MyUniBroDavis]

Noise. That is the term stat people like to use to explain away a data point they don’t like because it goes against their narrative. It’s not a sign the stat is junk, no not at all, couldn’t be. It must just be noise because the other data points of that stat fit with their narrative.

This has been today’s lesson on how stats are used subjectively to form an argument and are no more objective than the greatly mocked and horribly offensive “eye test” (the artist formerly known as watching the actual game of basketball).

who are you talking to?

The world.

Happy New Years.

I think it’s true the eye test isn’t used enough and it is the most important thing if you can be objective about it

That doesn’t mean stats themselves are wrong, noise is real and most of the criticism against statistics are more criticism against stats for people without critical thinking skills, or from people who look at stats without critical thinking skills

[OhayoKD]

who are you talking to?

The world.

Happy New Years.

I think it’s true the eye test isn’t used enough and it is the most important thing if you can be objective about it

That doesn’t mean stats themselves are wrong, noise is real and most of the criticism against statistics are more criticism against stats for people without critical thinking skills

Eh, it has limited utility if your priors are off, and priors that are constantly disagreeing with the best available data are probably off.

Good data-analysis -> good eyetest -> better data-analysis -> better eyetest