Prof: Stats don't predict head-to-head outcomes

Updated: December 12, 2003, 2:15 PM ET
Associated Press

LOS ANGELES -- Some Southern California professors who specialize in statistics are crying foul over the computerized ranking system that left the Trojans out of the national championship game.

Started in 1998, the Bowl Championship Series uses computer calculations of game statistics and outcomes to pick two teams to play for the national title each year.

James R. Beniger, an associate professor at USC's Annenberg School of Communications, said it's better to decide which team is the best on the field -- not through the statistics.

"As a statistician, I could find statistics to prove that USC is the best team in the country, and I could just as easily find some to show that another team is," said Beniger, who holds a master's degree in statistics from California.

Beniger said statistics provide good data on the past performances of teams, but added: "When you want to know the outcome of one team beating another team, statistics really don't help."

USC finished the regular season ranked No. 1 in The Associated Press writers poll and USA Today/ESPN coaches poll, which are decided by human voters. But the Trojans didn't get the nod from BCS computers, which placed the team third behind Oklahoma and LSU.

USC will play Michigan in the Rose Bowl game in Pasadena on New Year's Day while Oklahoma faces LSU in the Sugar Bowl three days later.

USC, Oklahoma and LSU finished the season with one loss each. The Trojans could still claim a share of the national championship by finishing atop the final AP poll.

Seven BCS computers make their selections based on the same set of game statistics but give differing priorities to pieces of data.

Fans know little about how these computers make their decisions or who programs them. As it turns out, the computers are, for the most part, run by a handful of math whizzes who happen to be sports fans.

One is Peter Wolfe, a Harvard-educated infectious disease specialist in Los Angeles. He received an earful from colleagues who are USC alumni but stands by his system.

"The ratings are right," he said. "USC played a weaker schedule and lost to a worse team than LSU and Oklahoma. I understand people's emotions about it, but that doesn't change what the ratings are."

USC lost to California, Oklahoma was beaten by Kansas State and LSU fell to Florida.

Wolfe, whose system uses a "maximum likelihood estimate" of performance, said he checks his scores with two other BCS computer operators and would allow any credible, independent person to review his numbers.

Geert Ridder, a USC professor of economics who has studied statistics in sports, takes issue with the "maximum likelihood estimate."

Although the system is a good tool, computer operators can't validate their models without a playoff of the best teams, Ridder said.

"If the men who make these computer models take themselves seriously they should be pressing the BCS to have a playoff," Ridder said. "Only by confronting predictions with outcomes can we hope to weed out the bad models and rankings."

Ridder says he's not a big USC football fan. A native of the Netherlands, he prefers soccer.

"My perspective of the BCS and the computer ranking is not clouded by a fog of anger, as seems to be the case for many of my colleagues," he said.


Copyright 2003 by The Associated Press