Breaking down Wins Produced

To paraphrase John Von Neumann,truth is much too complicated to allow anything but approximations.

Case in point, it is a simple truth that everything in the Wins Produced Model comes from the Boxscore. All the math is based on actual measurable events that occur and are measure on a basketball court. If you understand this it can lead to some interesting revelations about the game of basketball.

But that’s really getting ahead of myself. Let’s roll it back and start from the beginning because you see people always get confused with the model itself. One of the key misconceptions is that we somehow make up some of the numbers that actually go into the model itself. We always get asked this question. This is patently not true but I can actually understand how people actually get this wrong.

Bear with me

Now Prof. Berri has a wonderful page outlining in detail how the model is actually built. This is fabulous is you want to replicate how the model is built and you want to go off and do it yourself. It’s actually, to me, the easiest of the advanced models to build and replicate (and trust me I’ve done them all). The problem lies in how faint that praise really is.

The level of detail in that explanation can be very intimidating to the more casual fan. I must admit dear reader that even I can get confused and lost at the intricacies of the model.

Trust me to take it upon myself to attempt to clarify. My goal then here is to walk you thru the model and clarify some common misconceptions that exist.This will then allow me to have some fun with the components of the model.



The first bit is simple. Wins Produced is based on point margin and the fact that it correlates strongly to winning (Rocket Science, I know). Points scored on offense and given up on defense are good indicators of a teams ability to win games.On this pretty much everyone agrees.

In the simplest terms, there are really four things that actually matter to winning in basketball:

  • Points Scored on offense
  • Possessions Employed on offense
  • Points allowed on defense
  • Possessions Acquired on defense

What the model actually does is map every action on the court that’s measured in the boxscore to one of those four things and map out it’s value to wins.

The model then is divided into three distinct parts:

  • The Individual player’s production on the offensive end. This includes Field goals made and missed, Free throws made and missed as well as turnovers and assists or all the player actions related to the end of his teams play on the offensive end. Let’s call this Individual Offense.
  • The Individual player’s production on the defensive end and on the Boards. Here we have all the plays related to a player gaining his team possession of the basketball by ending or preventing the opponent’s offensive opportunities this includes rebounds (offensive and defensive), steals and blocks. Let’s call this Individual Defense + Rebounding .
  • The team factors. These capture all those boxscore stats that are specifically not assigned to an individual. The first group includes the rebounds and turnovers that are not assigned. The second includes all the missed field goals for the opponent not captured in the individual stats via blocks. Let’s call this Team Defense.

Now the majority of these stats are just tallied up and multiplied by a value to come up with out total productivity. There are three exceptions to this. Team Defense factors are divided equally amongst the players on a team based on minute played by each player. In essence, this gives you credit for playing on a good or a bad defensive team. Assists are the second exception. In the simplest possible terms, the player gets credit for assists in comparison to an average player on his team. Better than average assist producers get credit, below average guys get docked (and nobody likes playing with them). The assist itself get assigned about half the value of the points produced by the field goal assist. The last exception is defensive rebounding. For each defensive rebound gained by a player he takes away about half a rebound from his teammates. What we then do is take the value of that half a rebound, toss it in a rebound tip jar which gets spread evenly amongst the teams player’s based on minutes played.

So a player’s individual productivity is the sum of Individual Offense, Individual Defense and Rebounding. This is then adjusted by the value of Team Defense to give total player productivity. To get Wins Produced, we compare the player’s (or team’s) productivity versus the average productivity at his position (or for the average team). A player’s value is then a function of how good he is versus the average opponent faced. The classic model does this in Wins but we can also very easily translate it to points (and I’ve done just that).

If this is too much trouble for you, we’ve even come up with a nice simple version:

Win Score =  PTS + STL + ORB + 0.5*DRB + 0.5*AST + 0.5*BLK – TOV – FGA –0.5*FTA – 0.5*PF

The data set for all the work is here (Team Data)

Let’s get to the payoff. If I work the numbers for all these factors since 1978 I come up with some very interesting results. First let’s look at last year by category as a proof of concept:

(Editor’s Note: I screwed this up initially, fixed now. Carry on.)

Chicago,Miami, Philadelphia and OKC show up as our top 4 (I feel sorry for the Bulls fans). Boston, New York then Memphis show up as our best D’s (with the Celts being historically good). On offense the top three are San Antonio, OKC and Denver. For individual defense and rebounding we get Chicago, the Lakers and OKC. This lines up with reality very well.  The Lakers for example have been good at the possession acquired game in the last decade (offense and team d have always been a concern as well). The bottom of that list is fairly accurate as well.

Let’s get back to the historical chart though:

A few things jump out here:

  • The level of correlation for the total Wins Produced model to Wins remain very consistent over time in the 90%+ range.
  • The models containing just the boxscore stats (Wins Produced and Win Score) track very closely to each other.
  • The explanatory value of each of the factors changes dramatically over time. Offense has risen over time (hello three point shot) vs Individual defense and Rebounding but it has made a comeback more than a few times. Team defense in particular seems to go in an out of fashion (and if I had to guess it is very,very sensitive to rule changes).
  • Efficient Offense seems to be priority one to winning. Individual Defense and rebounding and Team Defense seem to be more arbitrary (good to have but no guarantee of success).

I hope this clarifies some doubts and leads to some more interesting questions. It’s certainly going to lead to some more post (yes, dear readers, that is foreshadowing).




Comments are closed.