« More articles in Uncategorized   |   Go Home

Lots of responses, so let’s clear some things up. First off, I may not have made my overall point as clear as I could have. Yes, diminishing returns do exist (of course), and should be accounted for when building a roster. But when we use stats to evaluate players, they should be predictive, not explanatory. In other words, we should only be using stats that will give us an impression of how proficient each player is, and this should be relatively constant no matter his surroundings. If a player’s stats are drastically different depending on his situation, we’re either looking at the wrong stats, or looking at them wrong.

Imagine evaluating a baseball player’s offensive rate stats without taking into account park factors. This would be heresy for most people who read this site. And while many defensive systems take into account park factors, the relative handedness and flyball/groundball tendencies of a team’s pitching staff, and a whole host of other factors, they clearly don’t consider the impact of the surrounding fielders enough.

Berri and Schmidt tried to use the law diminishing returns to explain why players’ numbers varied in certain cases, based on their “Wins Produced” stat. But that in itself disqualifies it as a predictive stat, and therefore isn’t of all that much use.

On a somewhat related topic, I got this response from a reader:

…it’s important to remember that the Berri stats measure efficiency more than anything. So even though Kerr’s stats are through the roof, that doesn’t mean he wasn’t very efficient at what he did. Just because Kobe has a bigger role, doesn’t mean that he couldn’t benefit from being more efficient at what he does.

- Griffin

The point I had made to him previously was that Wins Produced, or its offspring WP48, measures a player’s efficiency within his own role. The offensive ratings on Basketball-Reference.com estimate how many points that player produces per 100 possessions. Steve Kerr had a career rate of 122.1, the highest of all-time. His oft-forgotten teammate, Michael Jordan, scored in at 118, and Kobe Bryant is currently at 112.

This actually makes a bit of sense, once you realize what this rating means. Steve Kerr was asked to perform a certain offensive role, mostly hitting open jumpshots. He filled his role better than any player in NBA history. But now imagine if Kerr was to take on Jordan’s role, taking 30+ shots per game.

Long story short, you can compare Steve Kerr to Tim Legler (117), but not Hakeem Olajuwon (108). And you can compare Michael Jordan to Kobe Bryant, but not Manu Ginobili (113).

I’m not sure how well Basketball-Reference.com’s numbers correlate to Wins Produced, but B-Ref’s seem to be more conservative all around. I’m not up enough on basketball research to really delve into this, but if someone else is, I’d love to hear from you. Do these stats better adjust for context? And has there been any work done in adjusting for the player’s role? I’d be very interested.

Feedback? Write a comment, or e-mail the author at shawn(AT)squawkingbaseball.com

2 Existing Comments

Add New Comment

  1. on October 18th at 02:29 am
    Griffin Caprio said:

    You hit the nail on the head. These statistics ( really all statistics ) are meant to be used as input to decisions, not the decision makers themselves. Comparing Jordan to Kerr is like for different roles is like comparing a screw driver to a hammer when looking for a tool to drive a nail just because they’re both “tools”.

    It will be difficult to find predictive stats that are worth anything because sports are so erratic. You’ll essentially end up making up numbers that gravitate towards supporting a personal perception. Ask anyone who’s tried to come up with revenue projects for a new startup.

  2. on October 18th at 03:23 am
    Griffin Caprio said:

    That should be revenue projections, not projects.