More OR and Sports

I missed this earlier article from the Wall Street Journal entitled “The NBA Tries to Make Teamwork a Science” on how teams in the National Basketball Association are trying to measure “team” effects of their plays (thanks Otis Smith for pointing this out). Basketball, more than many sports, relies on smooth teamwork for a team to be successful. This leads to an interesting data mining issue. The article link expires in 7 days, but here are a few excerpts:

In a league long dominated by high-flying superstars, more teams are focusing this season on teamwork — and turning to surprisingly scientific methods to measure it. New technology makes it easier to track the performance of every combination of five players that steps on the court, in a long list of game situations, from out-of-bounds plays to pick-and-rolls to zone defenses. As different player mixes yield different results, teams are beginning to quantify the elusive concept known as chemistry.

[Wolf Pack: Eddie Griffin (far left) and Kevin Garnett (center) click on the court.]
Wolf Pack: Eddie Griffin (far left) and Kevin Garnett (center) click on the court.

Say, for example, that after a coach inserts two particular players into a game, the opposing team has trouble scoring. Getting ready for the next opponent, the coach might flip open his laptop, punch a few keys, and see how his team did defensively in other games when the same two players were on the court together. He’s able to do this because teams are increasingly turning to software that dissects plays, follows every pass and shot and tracks each player’s part in every possession.

Not surprisingly, it is people in Operations Research that come to the rescue. Wayne Winston is best known for the textbooks he has published on a range of operations research-related topics. But he also has been working with NBA teams on the issue of teamwork:

Over the course of an NBA season, the average team uses about 500 different five-man lineups, according to statistician Wayne Winston. The new push toward more number-crunching analysis could have profound implications for how games are played.

This issue clearly is important to teams:

Of course, the idea that a good player isn’t the same thing as a good teammate is probably as old as the NBA itself. And duos that just seem to click on the court aren’t a new phenomenon; you might remember some fellows named Stockton and Malone or Cousy and Russell.

But technology is taking the guesswork out of finding player combinations that work — or don’t. Shot-location data, for example, have often confirmed teams’ hunches that pairing guys who like to shoot from the same part of the court can inhibit both players’ scoring.

And teams that have bad chemistry often see it blow up in their faces. Exhibit A: The 2003-04 Los Angeles Lakers, who added future Hall of Famers Gary Payton and Karl Malone to an already stacked roster — and got worse. That team was hurt by injuries and the distraction of Mr. Bryant’s legal troubles, but Mr. Payton’s inability to accept a reduced role and play a team game was widely seen as a big part of the team’s ultimate undoing.

The Miami Heat took a major roll of the dice when it this off-season jettisoned Eddie Jones and Damon Jones, perhaps the two best teammates stars Shaquille O’Neal and Dwayne Wade could ask for, and replaced them with two highly skilled players — Mr. Payton and Antoine Walker — who have had trouble fitting in on equally talented teams. The Heat didn’t return phone calls seeking comment.

Science suggests the pairing of dominant scorers like the Philadelphia 76ers’ Allen Iverson and Chris Webber is likely to run into problems. Statisticians Jeff Sagarin and Mr. Winston created a calculator that rates the performance of the many different lineups each team uses during a season and found that one of the lineups the 76ers used most after Mr. Webber was acquired last year was more than 21 points per game worse than an average NBA lineup against similar competition.

So we see the statistics. An interesting research question in the OR tradition is to create a model that can explain what is going on, and allow predictions. If Eddie Griffen makes superstar Kevin Garnett better, how could that have been predicted beforehand in order to say, trade for the best “fit”?

Wayne Winston will be one of the speakers in a session of mine in this fall’s INFORMS Conference in Pittsburgh.

2 thoughts on “More OR and Sports”

Leave a Reply

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