Thursday, October 18, 2007

Larry Miller, stat man

Apparently, Larry Miller has his own handy back of the envelope way to evaluate whether his players had a good night:
It's pretty simple: Add up points, rebounds, steals, blocks and assists. Subtract fouls, turnovers and shots taken. Then divide by minutes played. This gives you what Miller calls a player's batting average, to borrow from baseball.
A player with a batting average of .300 usually approaches All-Star status, Miller said. A player with a .400 average is definitely an All-Star and a player with a .500 average is a likely MVP candidate. Karl Malone's career average, which Miller cited, was .457.
Over the years, it regularly spit out John Stockton as the NBA's top point guard, Michael Jordan as the top shooting guard, Larry Bird as the top small forward, Malone as the top power forward and Shaquille O'Neal as the top center, Miller said.
It's presented in a slightly different format then other stat measures, but eye-balling it I see why it seems to do pretty well. Moving the pieces around a bit, you can see that points minus shots taken serves as a crude measure of scoring efficiency, steals plus rebounds minus turnovers gives a crude measure of how many extra possessions the player gains (or loses) for the team, and blocks plus assists minus fouls gives a basic measure of the player's offensive and defensive help ability. Miller's formula is actually very close to the Win Score formula.

All of these stats are also kept for a team, so do they serve as a good proxy for team performance? The short answer (based on quick look at last season's numbers): Somewhat.

The top sixteen teams based on this metric were, in order: (1) Phoenix (with a team average of .288), (2) Denver, (3) San Antonio, (4) Golden State, (5) Utah, (6) Detroit, (7) LA Lakers, (8) Dallas, (9) Washington, (10) Toronto, (11) Chicago, (12) Houston, (13) LA Clippers, (14) New Jersey, (15) Sacramento, (16) Memphis (.230).

This list includes all eight Western Conference playoff teams. It also includes five of the Eastern Conference playoff teams. Of the excluded Eastern teams, Cleveland and Miami are the next two teams on the list. If you took eight teams from each conference using this measure for last season, the only playoff team that would not make the cut would be Orlando (they'd be beaten out by our Sixers!). So as an off-the-cuff measure of who's good versus who's bad, I'd say Miller's "batting average" does a decent job.

That said, it definitely suffers from the bias that most non-possession based systems suffer from-- running teams rank higher, even if they actually aren't better. With more possessions, there are simply more opportunities to get the statistics valued by this measure. Memphis had only 22 wins last season, but in the second half of the season they played at a very high pace. Good offensive teams also benefit from this system relative to good defensive teams-- thus Golden State, Denver, the LA Lakers, and the Wizards are ranked higher than I would expect form a neutral system, while the Rockets are ranked lower. And the Cavaliers, who made it to the NBA Finals last year on the back of a fantastic defense, but a mediocre (at best) offense fall out of the top tier of teams altogether.

What's the takeaway from all of this? Well, I think it shows that the NBA has done a credible job in figuring out good statistics to track. They might not track the "best" statistics, but the statistics they do track have at least some bearing on how teams perform. The teams that do a better job acquiring the "good" stats tend to be better than the teams that end up with a ton of the "bad" stats. Put those stats into a formula that meets the common-sense test, and you'll probably end up with results that aren't too outlandish. Not an earth-shattering revelation, but always good to have this belief confirmed when you spend a lot of time looking at basketball statistics.

No comments: