On Wed, 9 Oct 2002, Dean Oliver wrote:
> A key philosophy I've worked on is that players can only stay the
> same or get less efficient as they take on a greater load of the
> offense, given their teammates. Basically, they have to take worse
Yup, this is what separates econometrics from most other branches of
statistics: heavy reliance on theory or assumptions, to determine the
resulting statistical procedures or interpretations.
> shots or make more mistakes. This concept has to be true to some
> degree, otherwise the Bulls should have ALWAYS given the ball to
> Jordan, every single time. Players have to become less efficient as
> they take more possessions or else you'd have players like Steve
> Kerr, shooting 50% from the 3pt line, who should always shoot the
Yes, two good concepts being used here: proof by contradiction. And the
principle of diminishing marginal returns. They HAVE to diminish.
Another example would be Artis Gilmore late in his career, when he was
shooting something like 65% from the field. Why didn't he just shoot the
ball 40 times per game? For the obvious reason that his FG% (and
efficiency) would've gone way down.
> There are players who, with a basic look at their numbers, don't
> support this. Iverson is one. His efficiency goes up with more
> possessions. Why? Because there are nights when a team tries to
> cover him 1 on 1 and he gets hot, so he goes all night killing the
> horrible defense. But such a look at offensive rating (points
> produced per 100 possessions vs possessions used) is mixing cause and
> effect. If a team were double teaming him and he kept trying to
> score, his efficiency would go down. Fundamentally, it is very hard
> to sort out cause and effect in looking at efficiency vs the load
> that players take on.
This is a good example, but is only one of the complications. It's also
an example of how we rely upon our theory or assumptions: in this case
the assumption that, all else being equal, an increase in possessions by
a player must result in a decrease in his efficiency. If we observe
some player's stats going in the opposite direction, such as Iverson, we
resort to explanations such as DeanO's.
A good procedure, but only as long as the theory or assumptions are good.
In this case, the assumption of diminishing marginal returns is a good
But there are yet other complications which make estimating the
possession/efficency tradeoff difficult. In addition to the defensive
context, there's the teammate context: what sort of teammates does the
player have, how good are they, do their skills complement each other
(Shaq & Kobe) or compete with each other (Bellamy & Willis Reed, Wilt &
Thurmond -- there was no point in having both of those players on one
team, the team was better off trading one of them away). During a season,
most of the teammates remain constant at least so we can try to use
individual game data to explore the relationship between possessions and
But then there's yet another complication, especially when we're looking
at a player's career across time: the quality of the player almost
certainly varies. Most players improve, plateau, and then decline. So
their possession/efficiency tradeoff is going to be varying over time.
It's like trying to estimate the relationship between how fast you drive
and what kind of gas mileage you get: but some days your car is a Ford
Excursion and some days it's a Ford Fiesta or whatever they call those
little cars. Fortunately career swings aren't completely random, there's
some regularity and stability to them.