The WOWY Thread

Moderator: Doctor MJ

ElGee
Assistant Coach
Posts: 4,041
And1: 1,202
Joined: Mar 08, 2010
Contact:

The WOWY Thread 

Post#1 » by ElGee » Wed Jul 9, 2014 8:27 pm

I used to have a section for WOWY (or in/out) on blog. Since then I've improved the method, controlling for other key players and when possible, for when opponent key players missed games as well. I've also introduced a metric designed to capture WOWY quality in a single number, called WOWY Score, which is detailed below. There is also now a WOWY spreadsheet linked below.

This thread will be a collection of WOWY reports related to key players in NBA history and will be updated over time. I've also finally added some formal statistical analysis around the variance/accuracy of these numbers, and that is included in a player's table.

Benefits of WOWY or In/Out
Spoiler:
1. It CORRECTS -- in a big way -- the general perception of "how a guy's team performed" with him

    This is perhaps the biggest impetus for examining what happened with and without a player. When people ballpark team value or players, they often have no idea what actually happened with the players in question in the lineup.

2. It fairly accurately demonstrates situational value, which is really important

    The major payoff is an attempt to "isolate" a player's SITUATIONAL VALUE. It is dependent on backups, team circumstance, etc. But nonetheless, this situational value is tremendously helpful in evaluating players. It is not simply an extension of on/off, because coaches cannot co-vary players with lineups within a game. Instead, we are observing something more "powerful" which is the observed effect with or without the player.

(Please note that the team structure must be considered. If Pau Gasol is backed up by a clone of Pau Gasol, then when Pau Gasol gets injured the team won't miss a beat. Similarly, if Kurt Thomas is the last big left on the 2006 Suns roster, when he gets injured, the team loses a lot.)

3. Even in smaller samples (e.g. 10g, 14g) it suggests trends that are likely

    It's hard to have really strong 20+ game sample sizes. However, even in small samples, as shown in the +/- tables, having a number of independent events that suggest a trend is valuable information. (This is also captured in the WOWY score itself.) Knowing how rare it is for a team to play +8 basketball over 8 games doesn't mean that a team in a given circumstance is +8, but it might mean that it's very likely that the team is an above-average team, which can be a critical new data point in an analysis.


Methodology
Spoiler:
PART I
The basic method for WOWY (with or without you) or in/out is to calculate team performance in games when a player played versus games when a player played. Since we are trying to isolate the player as much as possible, "controls" will be noted when applicable. For instance, if you see "Johnson In (70)", that means the in/out is ONLY across games where Johnson played, and that he played in 70 games. Non-Johnson games are excluded.

In addition, if you see multiple years runs (e.g. "68-70)") that run is almost always partitioned around a similar core group of players. Thus, if major roster changes took place, it's unlikely you will see a multiple year run that crossed over that period.


Error Rates/Accuracy
Spoiler:
I've sampled (not truly random, for the scientists in the room) games from 2008-2012 of different sample sizes and analyzed the variance. 100-team seasons were looked at (20 teams over 5 years) and for each sample size of games, 82-sample size were sampled. For instance, 74 8-game sample sizes were pulled for all teams over 5 seasons. 58 12-game sample sizes were pulled. Etc.

Regressing the results yields an equation (with R^2 .973) that can predict the variance of a typical sample-size of games in relation to the "true" SRS of a team. For samples under 60-games (where variance asymptotes):

    sample_std = 7.777 + (-1.857 * ln(sample_size))

It's worth noting that variance sort of asymptotes at around 0.45 standard deviations after 60 games. In other words, if we look at an 82-game sample and keep increasing the sample (i.e. the playoffs), SRS exhibits the same variance it does after about 60 games. Thus, any SRS, even at the end of the season, isn't "perfect" but it's 95% likely to reflect the "true" SRS of a team within +/- 1 point.

This has been factored into the 80% and 95% +/- intervals displayed in a player's table. The +/- figure shown is SRS points away from an 82-game sample (technically 0.46, or average SRS std after 82 games). Thus, the simplest way to think of the accuracy of the sample shown in either the 80% or 95% interval is "how many SRS points away from a stable, season-long sample."

PS It would be mildly more accurate, but sometimes even significantly more accurate, to incorporate the variance of the games sampled. Unfortunately, I haven't done that in the past and am not going to update and re-scoop 300+ WOWY runs.

PART II UPDATE:
I investigated whether or not teams are more consistent with a healthy (controlled) lineup. They indeed are, and regression on those numbers, (this time using truly random sampling of games!) yields the following more accurate formula than the one listed in Part I above-- again for samples under 60 games:

    sample_std = 7.4757 + (-1.8792 * ln(sample_size))

This is the formula currently used for error rates and for WOWY scores.


WOWY Score
Spoiler:
Given that I have error rates, I can now use this information to create a WOWY Score -- a single number that essentially captures how impressive a given WOWY run is. The number is calculated like this:

    1. Determine the SIO of the run. SIO is "Simple In/Out" and it slightly adjusts the difference between the in and the out value based on the quality of the team. Weaker teams are considered slightly easier to improve, quality teams harder. The exact formula can be seen in the post "Calculating Championship Odds."

    2. Once SIO is determined, use the 95% confidence interval to determine how likely the SIO score is. The lowest possible confidence interval (a 1-game sample) is +/- 15.1 points. Thus, if the confidence interval is +/- 7.5 points, the "likelihood" in this case is 50% toward total confidence. The formula is simple for this SIO adjustment:

    ((15-confidence_interval)/15) * SIO

    3. Once we've done that, there is a year adjustment based on how easy/hard it is to create separation in points in a given year. This is based on an analysis of the shot-clock era and the z-scores (variability) in different leagues. The adjustment was based on sampling the top teams in each league over a rolling number of years and normalizing z-scores to SRS points. The early 1970's are the easiest environment to create separation in SRS points, and the post-merger years the hardest.

The resulting number is WOWY score. WOWY score is designed to be a quick number to instantly interprets the quality of a WOWY run. WOWY scores can be seen in the WOWY spreadsheet.

Note: To calculate multi-season WOWY scores, the weighted average of the SIO was taken, then the weighted average of the yearly adjustment and finally any adjustment for confidence interval depending on the sample size of the entire run.


LATEST UPDATE: Robinson added 7/28/14 -- Note, some minor errors still present on the images below that are correct in the WOWY spreadsheet.

Here is the link to the WOWY spreadsheet: https://docs.google.com/spreadsheets/d/ ... sp=sharing (If you see a date format, that's google spreadsheet reading something like "04-05" as April 5th, or "04/05"." In that case, the month will be the first year of the sample and the day the second (e.g. "08/11/14" = "08-14.")

Abdul-Jabbar, Kareem
Spoiler:
Image


Barkley, Charles
Spoiler:
Image


Bird, Larry
Spoiler:
Image


Bryant, Kobe
Spoiler:
Image


Chamberlain, Wilt
Spoiler:
Image


Duncan, Tim
Spoiler:
Image


Erving, Julius
Spoiler:
Image


Garnett, Kevin
Spoiler:
Image


James, LeBron
Spoiler:
Image


Johnson, Magic
Spoiler:
Image


Jordan, Michael
Spoiler:
Image


Malone, Karl
Spoiler:
Image


Nowitzki, Dirk
Spoiler:
Image


Olajuwon, Hakeem
Spoiler:
NB: "Tomjanovich" should say "Chaney."
Image


O'Neal, Shaquille
Spoiler:
Image


Robertson, Oscar
Spoiler:
Image


Robinson, David
Spoiler:
Image


Russell, Bill
Spoiler:
Image


West, Jerry
Spoiler:
Image
Check out and discuss my book, now on Kindle! http://www.backpicks.com/thinking-basketball/
User avatar
bondom34
Retired Mod
Retired Mod
Posts: 66,580
And1: 50,199
Joined: Mar 01, 2013

Re: The WOWY Thread 

Post#2 » by bondom34 » Thu Jul 10, 2014 5:51 am

Awesome man, read the articles from your old blog, and can't wait to see the new results!
MyUniBroDavis wrote: he was like YALL PEOPLE WHO DOUBT ME WILL SEE YALLS STATS ARE WRONG I HAVE THE BIG BRAIN PLAYS MUCHO NASTY BIG BRAIN BIG CHUNGUS BRAIN YOU BOYS ON UR BBALL REFERENCE NO UNDERSTANDO
Doctor MJ
Senior Mod
Senior Mod
Posts: 50,604
And1: 19,355
Joined: Mar 10, 2005
Location: Cali
     

Re: The WOWY Thread 

Post#3 » by Doctor MJ » Sun Jul 20, 2014 5:41 pm

I echo what bondom said. This work is extremely useful.
Getting ready for the RealGM 100 on the PC Board

Come join the WNBA Board if you're a fan!
ElGee
Assistant Coach
Posts: 4,041
And1: 1,202
Joined: Mar 08, 2010
Contact:

Re: The WOWY Thread 

Post#4 » by ElGee » Sun Aug 3, 2014 12:00 am

I've added a link to a spreadsheet with all my current WOWY runs and also added a WOWY score, which is detailed in the OP.
Check out and discuss my book, now on Kindle! http://www.backpicks.com/thinking-basketball/
User avatar
bondom34
Retired Mod
Retired Mod
Posts: 66,580
And1: 50,199
Joined: Mar 01, 2013

Re: The WOWY Thread 

Post#5 » by bondom34 » Sun Aug 3, 2014 7:25 am

Fantastic analysis ElGee!
Just did a first glance, the 2 things that stood out personally were the general tendency for some people to overrate Rondo (I'm still not a big fan, and the numbers here seem to show it), and how sad Walton's injury problems really were.
MyUniBroDavis wrote: he was like YALL PEOPLE WHO DOUBT ME WILL SEE YALLS STATS ARE WRONG I HAVE THE BIG BRAIN PLAYS MUCHO NASTY BIG BRAIN BIG CHUNGUS BRAIN YOU BOYS ON UR BBALL REFERENCE NO UNDERSTANDO
lorak
Head Coach
Posts: 6,317
And1: 2,231
Joined: Nov 23, 2009

Re: The WOWY Thread 

Post#6 » by lorak » Sun Aug 3, 2014 9:21 am

ElGee wrote:I've updated the WOWY thread and created a spreadsheet with all of my current runs: https://docs.google.com/spreadsheets/d/ ... sp=sharing

There are still a few I need to transport over and I few more I need to run, but the current data set is fairly complete. I still think the easiest way to use this information analytically is to look at the year-by-year behaviors of a team (e.g. even if Superstar A misses 0 games, what else happens when starting center goes down for 25g?) You can use the spreadsheet to do this on your own, as I do not have time to make (and update) historical runs for players outside of the top-20 provided in the WOWY thread.


Why SIO results are different than the ones you posted some time ago? For example:

Code: Select all

player,year   ols G   old SIO   new G   new SIO
Baylor 1962    32   4,4   30   6,0
West 1968    31   7,0   28   8,3
West 1971    13   6,3   25   6,6
Bird 1991    22   8,0   22   9,4
Bird 1992    22   3,7   22   0,8
West 1969    21   5,1   21   5,2
Walton 1978    24   10,4   18   10,9
Robertson 1972    18   10,8   17   4,6
Walton 1977    17   9,1   17   9,6
Nash 2005-07    16   8,2   16   11,7
Barkley 1991    15   1,1   15   3,4
Baylor 1966    16   0,5   15   -2,5
Duncan 2005    16   10,3   13   10,2
Robertson 1970    11   4,4   13   4,9
West 1967    16   4,1   13   5,1
Barkley 1987    14   2,4   11   2,0
Barkley 1996    11   3,6   10   5,1
Barkley 1997    29   3,9   10   3,6
Magic 1988    10   8,2   10   10,3
Robertson 1968    21   3,0   10   9,0



Sometimes number of games is different, so it probably would explain difference (but how you decided which games to use?), but sometimes number of games is the same, but SIO different. "Controls" made so big difference, or you are using different method to calculate SIO?

PS
What happened to 1980 MIL Lanier? ;)
ElGee
Assistant Coach
Posts: 4,041
And1: 1,202
Joined: Mar 08, 2010
Contact:

Re: The WOWY Thread 

Post#7 » by ElGee » Sun Aug 3, 2014 8:46 pm

lorak wrote:
ElGee wrote:I've updated the WOWY thread and created a spreadsheet with all of my current runs: https://docs.google.com/spreadsheets/d/ ... sp=sharing

There are still a few I need to transport over and I few more I need to run, but the current data set is fairly complete. I still think the easiest way to use this information analytically is to look at the year-by-year behaviors of a team (e.g. even if Superstar A misses 0 games, what else happens when starting center goes down for 25g?) You can use the spreadsheet to do this on your own, as I do not have time to make (and update) historical runs for players outside of the top-20 provided in the WOWY thread.


Why SIO results are different than the ones you posted some time ago? For example:

Code: Select all

player,year   ols G   old SIO   new G   new SIO
Baylor 1962    32   4,4   30   6,0
West 1968    31   7,0   28   8,3
West 1971    13   6,3   25   6,6
Bird 1991    22   8,0   22   9,4
Bird 1992    22   3,7   22   0,8
West 1969    21   5,1   21   5,2
Walton 1978    24   10,4   18   10,9
Robertson 1972    18   10,8   17   4,6
Walton 1977    17   9,1   17   9,6
Nash 2005-07    16   8,2   16   11,7
Barkley 1991    15   1,1   15   3,4
Baylor 1966    16   0,5   15   -2,5
Duncan 2005    16   10,3   13   10,2
Robertson 1970    11   4,4   13   4,9
West 1967    16   4,1   13   5,1
Barkley 1987    14   2,4   11   2,0
Barkley 1996    11   3,6   10   5,1
Barkley 1997    29   3,9   10   3,6
Magic 1988    10   8,2   10   10,3
Robertson 1968    21   3,0   10   9,0



Sometimes number of games is different, so it probably would explain difference (but how you decided which games to use?), but sometimes number of games is the same, but SIO different. "Controls" made so big difference, or you are using different method to calculate SIO?

PS
What happened to 1980 MIL Lanier? ;)


You aren't the first to have this question to let me be explicit and hopefully that will clear things up. The decision to control for other players in the lineup is trying to control a confounder. For the most part, I'm controlling for players that averaged at least 25 mpg on the team that season. There are times when that might not hold true, but essentially that's the gist.

Controlling for players CAN make a huge difference, which is exactly why it's done. Why? Because let's say when Duncan misses games so too does Ginobili and Parker. That can have a massive impact. Also, that's exactly why we control, because saying

"Duncan out" (when Manu and Parker are out too) and comparing to "In" (when Manu and Parker are in) is a lot different than saying "Duncan out" (and Duncan alone) vs. Manu and Parker In. In general, we are going for:

A, B, C, D, E, F vs.
A, B, C, D, E

Where the key player is player F. This gives us much more valid and powerful results. Also, some old players need to be re-done (like 1980 Lanier) and I will update when that happens.
Check out and discuss my book, now on Kindle! http://www.backpicks.com/thinking-basketball/
DQuinn1575
Sixth Man
Posts: 1,932
And1: 705
Joined: Feb 20, 2014

Re: The WOWY Thread 

Post#8 » by DQuinn1575 » Sun Aug 3, 2014 10:27 pm

Great stuff-69 was

Any chance you can note trades - maybe in the control part - for example DeBusschere in 69 was basically replaced by Bellamy, in a trade, as opposed t being replaced by a sub - so it would probably get more credit (or Bellamy less)
ElGee
Assistant Coach
Posts: 4,041
And1: 1,202
Joined: Mar 08, 2010
Contact:

Re: The WOWY Thread 

Post#9 » by ElGee » Sun Aug 3, 2014 10:57 pm

DQuinn1575 wrote:Great stuff-69 was

Any chance you can note trades - maybe in the control part - for example DeBusschere in 69 was basically replaced by Bellamy, in a trade, as opposed t being replaced by a sub - so it would probably get more credit (or Bellamy less)


Yes I've had the exact same thought -- you will see trades mentioned in the control notes. It's not comprehensive right now, but I've been trying to add that lately.
Check out and discuss my book, now on Kindle! http://www.backpicks.com/thinking-basketball/
ElGee
Assistant Coach
Posts: 4,041
And1: 1,202
Joined: Mar 08, 2010
Contact:

Re: The WOWY Thread 

Post#10 » by ElGee » Mon Aug 4, 2014 5:41 am

UPDATE: There was an error with the 3rd tab "Prime Only" -- the WOWY score was calculating incorrectly. It is now correct.

If we believe that there is relative continuity in a player's performance throughout his "prime" -- not always the case, but let's pretend -- then the "Prime Only" tab is sort of like a value "footprint" that a player leaves over multiple years if he's kind enough to miss some games. Some issues of course are:

-the best players ever don't always miss sufficient games, especially in their peak years
-the score you see still only represents situational value
-even from the most consistent players, primes are rarely perfectly "consistent"

I am generally more impressed with positive scores in a larger variety of situations. (The larger the variety, the less "situational" the value is.) Playing with different teammates, different coaches, different systems, etc. and still moving the needle is perhaps the strongest signal of impact we can see in basketball. Looking at the current (08/03/2014) large but still incomplete data set, we see:

THE GOOD
-West looking like the King
-Oscar not far behind
-Steve Nash displaying (as advertised) value in Phoenix
-Nate Thurmond rounding out the top-5, and arguably one of the 5 most impressive results
-Shaq displaying (as advertised) value in 3 cities
-Robinson, Olajuwon, Duncan and Boston-Garnett excelling
-players like Pierce and Eddie Jones and McGrady quietly looking amazing

THE BAD
-Wilt sitting 39th (currently) from his 2 trade runs in 1965 and 1970
-Clyde Drexler -- 6 years, 108 missed games, 2 teams...+0.8 Wowy score
-McHale...how much of that is playing behind Bird
-Chris Webber...didn't really seem to matter
-Moncrief...the same
-Adrian Dantley -- looking like the opposite of West. An amazing 220 missed games and clear negative value for his teams.
Check out and discuss my book, now on Kindle! http://www.backpicks.com/thinking-basketball/
Chronz
Starter
Posts: 2,199
And1: 467
Joined: Jul 30, 2008

Re: The WOWY Thread 

Post#11 » by Chronz » Fri Nov 21, 2014 7:58 pm

Cant access team menu on the site, WHERE IS IT?

Return to Statistical Analysis