SIO: In/Out Data over Multiple Years

[ElGee]

For those who have followed along with last year's RPOY and this year's top 100, I've tried to provide data on changes in teams margins of victory (MOV) with and without a player in the lineup. (in/out) It's a crude process, and it's subject to all the issues of the +/- family, but it does answer the question "how did that team perform when the player was out?" It's more a measure of conditional value, but don't forget conditional value is often highly tied to absolute value at the superstar level...

As of now, I decided to crunch the numbers over multiple years for whatever I had in my library. This also opens the door to running all players games, using seasons they only missed a game or 2 as well.

After that, I've decided to make a simple adjustment for team strength by averaging the team's MOV with player's net difference when in and out of the lineup. The motivation behind this simple adjustment is to note that it's easier to improve weaker teams than it is to improve difficult teams, and that taking a 6 SRS team (championship contender) to 10 SRS (GOAT-level) is more difficult than taking a .500 team (0 SRS) to a 4 SRS team (52-wins on average and not really a contender).

So far, these are the results, sorted by the simple In/Out adjustment, or SIO (Simple In/Out) with a 29-game cutoff so Dr. J could be included:

Player Years Games MOV Net SIO
Walton 77-78 41 9.3 13.0 11.2
Nash 01, 05-07, 09 36 5.6 9.5 7.5
Duncan 00, 04, 05 37 8.4 6.5 7.5
King 84-85 33 1.8 11.2 6.5
McHale 86, 88, 91 45 7.8 4.3 6.0
Bird 91-92 44 6.3 5.8 6.0
Rodman 93, 95-97 100 7.8 3.5 5.6
Pippen 94, 98 48 7.6 3.3 5.4
Penny 97, 00 55 3.8 6.9 5.3
Garnett 06-11 72 5.7 4.9 5.3
Shaq 96-98, 00-04 142 6.4 4.1 5.3
West 67-69, 71 76 4.7 5.8 5.3
Hakeem 86, 91-92, 95-96 72 3.5 6.3 4.9
Kareem 75, 78 37 3.1 6.7 4.9
Mourning 94, 96-98 74 4.1 5.2 4.7
McGrady 02-04 28 -0.6 9.9 4.7
KJ 90, 93-97 129 4.7 3.7 4.2
Erving 73, 78, 83 29 4.1 4.2 4.1
Kidd 00, 04-05 46 2.9 4.9 3.9
Kobe 00, 04-07, 10 79 3.9 3.5 3.7
Barkley 87, 91, 94-97 100 3.5 3.5 3.5
Odom 05, 07 44 0.3 5.7 3.0
Cowens 75, 77 47 2.8 3.1 3.0
Pierce 07, 10 46 0.1 5.3 2.7
Ewing 87, 94-96 31 -1.1 6.4 2.6
Baylor 61-62, 66 54 2.3 2.4 2.4
Drexler 90, 93, 94, 96 90 2.4 1.0 1.7
Moses 78, 84 36 -1.1 4.2 1.6
Iverson 00-02, 04, 06 89 0.5 2.7 1.6
Webber 95, 97-98, 01-03 104 2.5 0.7 1.6
Wilkins 92-93 51 -0.3 3.1 1.4
Allen 02, 04, 07 66 -0.7 2.6 0.9
Hill 95, 00, 05 35 -2.5 4.2 0.9
Wade 04-08 95 -1.5 3.2 0.8
Wilt 65, 65, 70 156 -0.3 0.8 0.3
Paul 07, 10 55 -1.6 1.2 -0.2



Apologies for the crudity in presentation...I am planning a neater blog about this idea at some point in the future.

[ronnymac2]

I posted this in the top 100 # 16 thread to nobody in particular, so ignore the possibly hostile tone.

These regular season on/off numbers- when players missed games or whatever- are nearly useless to me. I see value in APM, RAPM, etc., because that more clearly shows interaction with teammates over longer periods of time. We get whole seasons to see a player's value to a team. The plus/minus family can tell us something.

But this on/off stuff resulting from player injury where stars miss entire games...man, how can I believe anything these numbers say? Anything can happen in a few regular season games. In the regular season, you don't get to gameplan against a team like you do in the playoffs. If a bad team from the West is on a hot streak, they can still easily take out the best team in the East during a random regular season game for a multitude of reasons: The great Eastern team isn't motivated, a player gets hot, matchup issues (that could easily be remedied with postseason gameplanning), coaches not taking gambles with foul trouble, limiting minutes, a player injury that you don't hear about, the flu, a team playing way better on the road or at home, etc.

I'll also bring up an idea that I've talked about before. Some teams are better built to lose a star player over the grind of an 82-game regular season. Other teams aren't equipped to handle injuries, but if they have their irreplaceable components relatively healthy and set (a reasonable hope unless you're from Portland or something), they have a greater chance at winning in the NBA playoffs than the team equipped to lose its star in the REG SEA. It's about team construction. That doesn't necessarily make a player better or worse. It has to do with valuable to his respective team. These players aren't being put into the same situations (in terms of neither talent nor fit), so looking at their value this way doesn't make much sense when deciding who the BEST players are.

Regular season results don't matter that much in the end. The regular season is for a team's habit formation and familiarity; the playoffs give us the best indication of which team is the best (not a be-all-end-all, but it's what we have).

I don't want you to get the wrong impression here- I do appreciate the work.

It's just that the disconnect I see between the player and his team makes me wonder what exactly these numbers are saying. I don't think this has the same problems as APM or any of the plus/minus models, because at least those models show us the value of a player through his interaction with his teammates during games over long periods of time. I'm not sure if this does.

[ElGee]

Hmmm - what do you mean by disconnect between player and his team?

The few 2003- seasons I've looked at have a decent correlation to on/off anyway...because this is a form of on/off. It's also data pre-2003, so I'm not sure what you're getting at...

To your main points
(1) Increasing the sample size is designed to avoid that variance -- and +/- family stats need huge samples to avoid variance, but consider that the 79 games Kobe missed in the years presented above represent roughly 3800 minutes of the Lakers playing without him. That's about 4 years worth of on/off/APM data...

(2) Some teams are better built to have their star on the bench. Period. That's why it's a measure of conditional value, not absolute impact. I'm not sure why this is a sticking point for you.

(3) Diminishing the regular season is wrong to me. It's highly predictive of the PS and what do you want to do, throw out every stat because of the "RS?" Just seems like such a strange view to have...then you'd have almost no tools to evaluate players and teams.

[Dipper 13]

What does this show for Big O & the Royals/Bucks in '68 (65g), '70 (69g), & '72 (64g)?

[ronnymac2]

Disconnect because he's not there during the game. That changes everything.

A team doesn't function the same way for the 10 minutes per game they are without their 38 MPG superstar as they do for 48 minutes when that player misses a game.


In a random regular season game, when the opposition really doesn't have the time to key in on how to properly take advantage of your missing star (at least to the fullest extent that it can), a team with talent can still kick ass even without its star. It can shift gears for a certain number of games and still play successful basketball against regular season teams, one game by one game. I know you adjusted everything to make it so raising a good team to the level of being truly elite team gets its proper weight, but...

Question- What conclusions do YOU draw from these numbers? Maybe that will clear things up for me. Maybe I'm thinking this is telling something different than what it's supposed to.

[ElGee]

I really don't understand why you have such a bias for off the court for 10 minutes vs. off the court for 48.

What's the evidence for a team not functioning the same? Do they try and play worse when the star is resting for 4 minutes? In other words, you may *psychologically* associate it as different game-planning, but 10 minutes off the court is 10 minutes off the court, whether he's coming back or not.

If you're suggesting that stars always rest in the same strategic period and that somehow changes things, I'm not sure how much I buy it. The other team can play (and does) any lineup during that period. As I've said before, on/off and in/out are two different pieces of the same pie -- why be so partial to one? All 82games on/off include missed games...and that when the missed games increase to a sufficient sample the correlation to in/out is incredibly strong.

For me, I think this first answers the question "how did those teams perform without the player" That can be a bit too results-oriented for me, so I'm always interpreting and weighing it based on sample/number of years, or trying to go deeper and understand context. Like all raw +/-, it is without a doubt a ballpark figure. And that figure is one of conditional value to a team...but the correlation between value and impact in a vacuum is quite strong. It's rare my MVP vote isn't a top-5 player and it's rare top-5 players won't be close to my MVP vote.

And if you're thinking "well, stuff like backup and depth matter," they matter in on/off too! That's the point. I've created the SIO adjustment for a quick snapshot of what kind of value a player was having on his team when we don't have on/off. It's a crude form of on/off then, or a slightly different form...and if I value on/off numbers a little or they make me rethink stuff, then I'd say the exact same for in/out/SIO, but with slightly lesser certainty/accuracy.

But again, note that ITO of variance, sample sizes are much BETTER in my in/out runs and especially the long-term SIO posted above. For instance, Derrick Rose sat 937 minutes this year. That's 20 full games.

[ronnymac2]

Responding to the first portion...Take the Bulls this season. Rose is the offense. When he isn't in the game, Chicago doesn't have a creator on the floor, but they stay in games because they're reserves come in and play balls-out defense for 11 minutes. Rose and a solid defense carries the team the other 37 minutes.

If Rose isn't in for 48 minutes of a game, those reserves aren't playing 11 minutes of balls-out defense. Chicago's excellent defensive players need to focus more on offense, which means they won't play defense as well. Their offense will also yield a lesser result than the offense with Rose.

The 10 minutes a star is off the court because of rest is not the same as the 48 minutes a star is off the court because of injury which prevents him from playing in the game at all. The fact that plus/minus doesn't distinguish that does troubles me a bit.


As for your conclusions...all right, no problems there. Definitely agree with the snippet about value and impact in a vacuum.