Yay Points! Boo Rebounds! What Gets You Played in the NBA?

Editor’s note: All of this is the great work of Ari Caroline, who you’ll note has been added to the about page! I’ve added some minor and organized the charts and tables. In short, any grammatical or commentary errors are my fault! The charts, tables and analysis section are the excellent work of Ari.

The intersection of advanced stats and conventional wisdom.

What makes a good coach? Is it inspiring the troops, grooming young players for succes, keeping locker room harmony? No! It’s playing the players that win. At least, that’s our opinion. However, it turns out things other than winning dictate things like pay, draft position and yes, even playing time in the NBA! Playing time, it turns out, wil be the focus of today’s post. We have a lot to cover. Let’s get started.

What Does it all mean?

We’ve got a lot of data to give you and sometimes that can be hard to understand. Luckily Ari has prepared a handy lookup table you can use!

Term Plain English Visual Explanation Value Explanation Position Breakdown for WP48
p-value Is this real?? Is the slope undeniably real? Look at the dark grey bands around the trend line (confidence of fit). Can you draw a flat line through the grey from one side to the other? If so, it’s not very convincing. Traditional approach looks for values< 0.05 In reality, depending on the number of variables, we usually look for a value < 0.01 C: Weak
PG: Strong!!
PF: Weak!!
SG: Middle
SF: Strong
R2 How much of the story does it tell? How tightly are the dots bunched around the trend line? A quick approximation is the light grey bands (confidence of prediction) and the RMSE measure in the upper-right corner (lower is better). There are no strict criteria for this. Even a low R2 can be very real, just a small part of the story. That said, if you have an overall R2> 50%, you’re explaining a fair bit. C: Weak
PG: Strong
PF: Weak!!
SG: Middle
SF: Middle
Slope Coefficient How dramatic is the effect? How steep is the slope? A steeper slope implies a stronger impact. Finance folks will recognize this as the beta. Caution: This is easily manipulated by changing the scale. Numbers are meaningless without the context. C: Middle
PG: Steep!!
PF: Middle
SG: Steep
SF: Steep

Ok, now let’s hop into the data!

What Gets You Played? The Cold Hard Numbers

2012 NBA Data: What Explains Playing Time?

Position Predictor p-value Incremental R2 Total R2
Centers PTS48 <0.001 42.1% 55.0%
AST48 0.003 8.0%
WP48 0.012 4.9%
Point Guards WP48 <0.001 45.0% 54.0%
PTS48 <0.001 9.0%
Power Forwards PTS48 <0.001 40.2% 46.3%
AST48 0.005 6.1%
Shooting Guards PTS48 <0.001 36.9% 44.7%
WP48 0.003 7.8%
Swing Forwards PTS48 <0.001 36.4% 43.0%
WP48 0.009 6.6%

For those more visually inclined, here’s the same data in graphical format.

  1. We stated before that WP48 is far more predictive of playing time for some positions than others
    1. Exhibit #1: MpG vs WP48 (Position Comparison)
    2. Clearly the relationship between WP48 and playing time is very convincing for Point Guards, less so for Shooting Guards, Swing Forwards and Centers and completely unconvincing for Power Forwards
  2. If the value that coaches place on overall productivity (as measured by WP48) in allocating minutes is limited, what do they, in fact, truly value?
    1. Duh, Points!: Exhibit #2: MpG vs PTS48 (Position Comparison)
      1. Using the primer above, we see a convincing relationship for all 5 positions
      2. Look how small the prediction band is for PFs!
    2. But not only Points, surprisingly, Assists!:  Exhibit #3: MpG vs AST48 (Position Comparison)
      1. There is a clearly significant relationship for 4 out of the 5 positions, including Centers and Power Forwards!
        1. Even though Centers only vary from about 0-5 AST48, it seems to effect their playing time significantly. Joakim Noah redemption!?
        2. It’s really sad that Shooting Guards are the exception here.
    3. Boo Rebounds: Exhibit #4: MpG vs. REB48 (Position Comparison)
      1. Only position that is even close to significant (but still not) here is Center
      2. Not Power Forward (this makes no sense at all)
      3. Or any of the other positions (predictable, but still sad)
      4. One could argue that Rebounds are included in the value placed in overall productivity (WP48).  Still, that is pretty minimal from an overall weight perspective since WP48 is secondary (or tertiary) for all but PGs and the algorithm includes so many other stats as well.
  3. Notes for Nerds
    1. I did look at some interactions and transformations, primarily to see if I could find anything that made Rebounds significant.  Nothing doing.
    2. The interaction of PTS48*AST48 actually supplanted the individual PTS48 and AST48 variables for Centers and PFs and increased the R2 a bit. However, I decided to leave it out for the sake of simplicity.
  4. Next Steps: Discuss the implications of this analysis for potential lineup choices and a really cool experiment that I’m trying to get my 8th grade daughter to do for the CT State Science Fair.


Comments are closed.