Defense Adjust Wins Produced(1st Pass)

This is meant as a quick post. I just took a quick pass at calculating Wins Produced adjusted by Individual defense.The main driver for this post is that David Lee is a bad defensive player and his wins produced are inflated

All numbers are taken from 82

First off some math explanations:

Full detail here

WP48 = PlayersProductivity(PROD48)– MATE48(Team Blk & Asst Adjustment)-DEFTM48 (Defensive Team Adjust)-RelAdjP48(Average of the first three terms for players playing the same position) +.099

The main term that could be said to drive defensive error is RelAdjP48 (the position adjustment). This term assumes that the player in question is facing average production at his position every night. This is very obviously not the case. Good defensive players can tend to be undervalued in our analysis. By calculating the Prod48 for a players average opponent we should be able to adjust for a players good or bad defensive skills. Given the rush nature of this exercise, I used the Celtics  as a baseline to assure total wins produced remained the same. I ended up with an adjusted position adjustment equal to my calculated WP48 for the average opponent +.07. When I have the time to complete the entire league for 2009-2010, I will refine the adjustment

Here’s my adjusted table for a few players of Note. Of the players I looked at here’s the ten best defenders:

  1. Kevin Durant
  2. Dirk Nowitzki
  3. LeBron James
  4. Dwyane Wade
  5. Kobe Bryant
  6. Dwight Howard
  7. Carmelo Anthony
  8. Brandon Roy
  9. Gerald Wallace
  10. Derrick Rose

Of note, Pau Gasol is a perfectly average defender. Of the Players who might be  involved in transactions that I looked at:

Joe Johnson & Amare are both better than average defenders. As for David Lee? He’s the worst in that list above but the story is not quite that simple. 82games claims that he used 72% of the Knicks minutes at C and only 3% at PF. If I work out his opponent’s numbers by position:

PosAdj_P48                     Actual OppAdjP48 (PF)                    Actual OppAdjP48 (C)

David Lee 0.406                                   0.307                                                           0.519

He is clearly a much better defender as a power forward than as a Center. A sign and trade to a team with a true center (say the Phoenix Suns) would benefit him greatly.

Arturo Galletti Written by:


  1. Shawn Ryan
    July 7

    Wow, that’s uncanny. I began working on the very same subject also because of the charges memetic charge that Lee is a poor defender. I was just looking at each individual box score and had planned on comparing the offensive production of the main player that he was defending vs. that player’s average production. Got really busy with a math class though, so I haven’t been able to follow through as of yet. We’ll see. It’s rather labor intensive. You’re approach is probably better as it is more widely without pretty much building your own stats database.

  2. Shawn Ryan
    July 7

    Wow, this is extremely interesting. A lot of WP48 darlings don’t look to good after this adjustment. Good job Arturo, you’re a huge asset to the community.

    You’re blog is definitely in my RSS feed after reading this post!

    • That’s pretty cool. I’m going to see if I can write a data mining script for this page.

  3. Shawn,
    I’m trying to see if I can build a tool to do this based on publicly available data. I can do a player in about a minute and a half at this point. I figure one season is about 12.5 hours (which is why I cherry picked). There is some interesting skew in the numbers and I think partly it’s due to the fact that offense in every regression I’ve done is more significant than defense. Once I have the full season I can incorporate some adjustments.

  4. Shawn Ryan
    July 8

    Yeah Arturo, keep up the good work. I’m planning on going through your regression post in detail soon. I think our interests on this stuff align closely, but I’m not very far into my computer science curriculum, so I don’t really have much chops. I think I’ll get acquire Minitab and see if I can, between that, your post and further reading get a handle on regressions.

    Really excited to see what you come up with though. Keep us updated.

  5. Shawn Ryan
    July 8

    Oh, and I don’t know how portable your tools are, but I can offer some data entry if any of that can be outsourced. You can email me if you’re interested. I think you should have my email since you use gmail and it saves the address of every single person that emails you ever.

  6. Shawn Ryan,
    It’s pretty portable (it’s an xlsx sheet). I’d be willing to e-mail it to you if you want to take a stab at it. Just pop me a note at my gmail account and i’ll send it to you.

  7. Shawn Ryan
    July 8

    Hmm… I suppose I was mistaken. I’d thought that I’d sent you an email before… perhaps it was a comment I was thinking of. Oh well. I guess I’ll have to give my email address to you here. You can send the xlsx to:

    shawn [dot] p [dot] r [at] gmail [dot] com

  8. Kevin Ferrigan
    July 13

    As someone who watched pretty much every game the Bulls played last year, I can tell you without question that Derrick Rose is not one of the top 10 defenders in the league. He’s pretty much rubbish on that end, actually. I suspect he was being credited in this system with the work done by his backcourt mate, Kirk Hinrich.

  9. merl
    August 23

    This post was awesome. I’d be really interested to see a followup to this, as this just seems flat out better than ‘basic’ WP.

  10. August 29

    This is very interesting but I think Kevin Ferrigan makes a good pt – this isn’t “defense adjusted” so much as “counterpart adjusted”. Because of cross-matches, no one can be sure if two opponents playing the same position actually guard each other. Maybe synergy sports data solves this problem but regular boxscores or play-by-play data don’t.

    • August 29

      It’s a better approximation. We have a project in progress to expand this in the future.

Leave a Reply

Your email address will not be published. Required fields are marked *