Do averages measure success?

[ronnymac2]

I have a question about averages.

Do you think averages measure success and failure properly? Or do you think they hide the truth effectively?

This goes for raw box score stats, plus/minus stats, hoopdata stats...basically, any stat you can think of that deals with averages.

Would a superior- or perhaps simply a complimentary- tool of analysis be looking at a distribution of player performance? So, average as well as median, mode, etc.

Let me know what you think. Maybe this thread branches out into something more.

[mysticbb]

High averages and low variance is what "superstars" usually have. There are a lot of more players capable of the same peak level performances, but they are most times way less consistent and have lower averages in the end. That makes looking at single peak level performances not a good way to measure a "superstar". Take Monta Ellis for example, the guy can easily go off for 40+ points or 30 points and 10 assists, but in the next game he will have 10 points, 5 assists and 8 turnovers. He is the prototype of an overhyped player due to peak level performance, but not with the necessary consistency in order to be a "true superstar".

In the end the averages are what we can expect from a player. Looking at things like variance, standard deviation or median tells us something about consistency and is indeed another really useful tool in order to determine "superstars".

[Rerisen]

Do you think averages measure success and failure properly? Or do you think they hide the truth effectively?

Isn't what you are asking for just splits? Maybe we just need good splits highlighted more. Performance in wins/losses, vs good teams and poor teams. Playoffs vs Regular season, do you step your game up, etc.

I would think that info is more valuable than just knowing a guy's high or lows divorced from anything else.

For example, Kobe's burning superhot nights. He has maybe half a dozen in his career, including the 81 points on the lowly Raps, and it gets highlighted as meaningful in lowbrow discussions, but just isn't really worth much in measuring him as a player, if you are talking about an occurrence that happens 5 or 6 times over a 16 year career, and pretty much completely at random. It's almost like a superstar version of the Monta Ellis effect Mystibb mentioned above.

[mysticbb]

Isn't what you are asking for just splits? Maybe we just need good splits highlighted more. Performance in wins/losses, vs good teams and poor teams. Playoffs vs Regular season, do you step your game up, etc.

We have to be careful about splits, because we are reducing the sample size and in most cases the overall averages are a better predictor than some splits. I like looking at the performance level of teams and players against "good" and "bad" teams (good means above average, bad below average) in order to put the stats into context. We see usually players perform better against bad teams, while worse against good teams, we also have a home bias. If the numbers are adjusted for this, they tell us more about it than simple averages. Well, that's why I included an adjustment for the strength of schedule in my boxscore metric.

It's almost like a superstar version of the Monta Ellis effect Mystibb mentioned above.

In a sense that is true. Bryant shows more variance than other players in terms of volume and efficiency. But he also was clearly better than league average in both anyway, that makes him still some sort of superstar.

[rrravenred]

Elgee did a series of posts somewhat around this topic.

It probably flatters to deceive in some cases (see Brandon Jennings' rookie season for example), but you'd need to unpick the context of each anomalous performance (good and bad) to really tease this out.

[ronnymac2]

Do you think averages measure success and failure properly? Or do you think they hide the truth effectively?

Isn't what you are asking for just splits? Maybe we just need good splits highlighted more. Performance in wins/losses, vs good teams and poor teams. Playoffs vs Regular season, do you step your game up, etc.

I would think that info is more valuable than just knowing a guy's high or lows divorced from anything else.

For example, Kobe's burning superhot nights. He has maybe half a dozen in his career, including the 81 points on the lowly Raps, and it gets highlighted as meaningful in lowbrow discussions, but just isn't really worth much in measuring him as a player, if you are talking about an occurrence that happens 5 or 6 times over a 16 year career, and pretty much completely at random. It's almost like a superstar version of the Monta Ellis effect Mystibb mentioned above.

Another I'd like to see is how a player performs against certain team types. So how does Kevin Love rebound against certain players/schemes/froncourts, what is the HoopData shot location info. for Russell Westbrook against big froncourts/quick frontcourts, etc. That tells us more about matchups in my opinion and could be a better tool for playoff predictions.

I guess it adds a bit too much subjectivity into the equation, but it's something.

[ronnymac2]

High averages and low variance is what "superstars" usually have. There are a lot of more players capable of the same peak level performances, but they are most times way less consistent and have lower averages in the end. That makes looking at single peak level performances not a good way to measure a "superstar". Take Monta Ellis for example, the guy can easily go off for 40+ points or 30 points and 10 assists, but in the next game he will have 10 points, 5 assists and 8 turnovers. He is the prototype of an overhyped player due to peak level performance, but not with the necessary consistency in order to be a "true superstar".

In the end the averages are what we can expect from a player. Looking at things like variance, standard deviation or median tells us something about consistency and is indeed another really useful tool in order to determine "superstars".

It really is all about consistency when it comes to superstars.

Question: When it comes to legitimate superstars (so don't take non-superstars/Monta Ellis-types into account), is it more valuable to be the most consistent relative to your peers (so low deviation with big numbers) or is it more valuable to be on the high-variance end of the superstar spectrum?

How much, if at all, does team construction matter when answering that question?

[EvanZ]

I recently calculated variance for RAPM for each of the past 4 seasons including this one. Here is a list of players (>1000 possessions in each season) sorted in ascending order by variance. You'll good players and bad players. For younger players, variance may just be a sign of rapid improvement (e.g. Durant, Jordan).

Code: Select all

FULL   MED   VAR
Dwyane Wade   2.33   0.04
David Lee   -0.52   0.05
Luke Ridnour   -0.48   0.05
LaMarcus Aldridge   1.48   0.07
Rodney Stuckey   0.17   0.09
Michael Beasley   -0.41   0.11
Mickael Pietrus   -0.32   0.13
Jason Thompson   -0.19   0.15
J.J. Redick   -0.07   0.15
Jared Dudley   0.15   0.16
Josh Smith   0.90   0.17
Raymond Felton   0.61   0.17
Kendrick Perkins   0.57   0.19
Paul Millsap   1.32   0.21
D.J. Augustin   -0.61   0.22
Andrew Bynum   0.84   0.24
Gerald Wallace   0.86   0.26
Marvin Williams   -0.47   0.27
Joe Johnson   0.88   0.28
Ronnie Brewer   0.25   0.29
Louis Williams   0.36   0.29
Chauncey Billups   0.87   0.31
Tony Parker   1.23   0.31
Leandro Barbosa   -0.40   0.31
Derek Fisher   1.80   0.31
Monta Ellis   -0.83   0.31
Andre Miller   1.66   0.32
Jose Calderon   -0.16   0.33
Danny Granger   0.98   0.34
Steve Nash   1.72   0.35
Matt Bonner   2.36   0.35
Nick Young   -0.17   0.37
Dwight Howard   2.28   0.38
Shane Battier   0.95   0.38
Pau Gasol   1.51   0.39
Thaddeus Young   1.09   0.40
C.J. Watson   -0.01   0.42
Amir Johnson   0.70   0.45
Kobe Bryant   1.61   0.46
Kevin Martin   -0.54   0.50
Tayshaun Prince   -0.42   0.50
Luis Scola   0.20   0.50
Trevor Ariza   -0.34   0.51
Dahntay Jones   -1.97   0.52
Roy Hibbert   0.22   0.52
Grant Hill   0.56   0.53
Richard Jefferson   0.51   0.54
Luol Deng   1.17   0.54
Rajon Rondo   1.26   0.55
Samuel Dalembert   -0.46   0.56
Daniel Gibson   0.47   0.57
Randy Foye   -1.08   0.58
Brandon Bass   -0.04   0.59
Marcus Camby   0.30   0.59
Russell Westbrook   0.59   0.61
Andre Iguodala   0.70   0.64
Jason Terry   0.97   0.64
Arron Afflalo   -0.22   0.66
Earl Watson   0.18   0.67
Courtney Lee   0.10   0.67
Daequan Cook   0.13   0.68
Jason Richardson   0.64   0.69
Nene Hilario   1.58   0.69
Nate Robinson   -0.13   0.71
Antawn Jamison   -0.38   0.74
David West   1.09   0.75
Ben Gordon   -0.41   0.76
Chris Paul   2.04   0.77
Carmelo Anthony   0.69   0.77
Drew Gooden   -1.78   0.78
Willie Green   -0.93   0.78
Devin Harris   -0.21   0.79
Tim Duncan   1.71   0.81
Chris Duhon   -0.62   0.82
Kurt Thomas   0.26   0.83
Metta World Peace   1.48   0.83
Rudy Fernandez   0.85   0.85
George Hill   -0.02   0.85
Jamal Crawford   0.47   0.86
Jarrett Jack   -0.95   0.86
O.J. Mayo   -0.15   0.86
C.J. Miles   0.27   0.87
LeBron James   2.73   0.89
Jason Maxiell   -1.66   0.89
Chris Kaman   -0.54   0.90
Stephen Jackson   -0.65   0.93
Joakim Noah   0.96   0.97
Jason Kidd   1.09   0.97
Kyle Korver   1.61   0.98
Chris Bosh   0.96   0.98
Beno Udrih   -0.42   1.04
Al Harrington   0.65   1.08
Emeka Okafor   0.06   1.10
Ryan Anderson   1.76   1.10
Glen Davis   -1.42   1.11
Rudy Gay   0.45   1.12
Boris Diaw   -0.12   1.15
Amare Stoudemire   0.33   1.16
Shawn Marion   -0.31   1.16
Al Jefferson   -0.13   1.18
Paul Pierce   1.29   1.19
Tony Allen   -0.18   1.20
Anthony Parker   0.21   1.20
Ray Allen   1.62   1.21
Hedo Turkoglu   1.05   1.21
Elton Brand   0.67   1.24
Mario Chalmers   0.69   1.28
Kevin Love   -0.03   1.29
Carlos Boozer   0.58   1.30
Marco Belinelli   0.15   1.33
Marc Gasol   0.68   1.39
Tyrus Thomas   -1.25   1.40
Dirk Nowitzki   1.55   1.43
Derrick Rose   0.79   1.47
Nicolas Batum   0.16   1.48
Mike Dunleavy   0.49   1.51
Vince Carter   1.65   1.58
Ramon Sessions   -0.38   1.59
Mike Conley   0.99   1.60
Kevin Garnett   2.59   1.64
Brandon Rush   -1.50   1.65
Caron Butler   0.66   1.69
Matt Barnes   -0.16   1.76
John Salmons   -0.86   1.76
Zaza Pachulia   -0.57   1.84
Steve Blake   -0.09   1.91
Marcin Gortat   -0.50   1.99
Jameer Nelson   1.31   2.04
Vladimir Radmanovic   0.83   2.16
Mo Williams   0.04   2.19
Kyle Lowry   1.09   2.32
Jared Jeffries   0.18   2.42
Damien Wilkins   -2.08   2.47
Marreese Speights   -1.36   2.51
Nick Collison   1.83   2.54
Rashard Lewis   0.30   2.64
Goran Dragic   -0.71   2.87
Tyson Chandler   0.52   2.90
JaVale McGee   -2.37   3.01
Chuck Hayes   0.04   3.02
Joel Anthony   -0.77   3.31
Anderson Varejao   0.93   3.37
Kevin Durant   1.15   3.63
Lamar Odom   1.09   3.82
J.J. Hickson   -1.90   3.86
DeAndre Jordan   -1.37   7.88
Channing Frye   0.79   10.57