After a spate of upsets broke open the NCAA tournament in 1986, including Cleveland State's victory over Indiana and Arkansas-Little Rock's win over Notre Dame, an NCAA basketball committee member told a reporter, "You can throw these computer ratings in the garbage." After reading that passage at home in Bloomington, Ind., Jeff Sagarin bolted for the phone. He called the committee member to point out that his ratings had Cleveland State 20th, even if the RPI ranked the Vikings 60 places lower. The member instantly apologized. "You're the only reason we took Cleveland State," he confessed to Sagarin. "They should have put you in uniform on their bench."
Better yet, the committee should put Sagarin in its conference room. Using a math degree from MIT and an M.B.A. in quantitative analysis from Indiana, he has designed the most widely respected system for gauging the relative strengths of teams and their schedules. And he's quick to credit the man who inspired a crucial part of his formula, the author of Sagarin's much-thumbed copy of The Rating of Chessplayers Past and Present—Arpad Elo, a Hungarian �migr� physicist who died in 1992.
After he began playing competitive chess, Elo noticed that it was possible to rise in the rankings despite losing and fall even if you won—precisely the anomaly that leaves so many fans baffled by the RPI. So in the 1950s he came up with a ranking system that foreclosed that possibility and struck a better balance between whom you played and how you did, always in light of expectations. Elo's system predicted almost exactly the number of games by which Bobby Fischer beat Boris Spassky in their 1972 face-off.
Click on the Men's Ratings link at usatoday.com/sports/sagarin, and you'll find three columns of information. One is Sagarin's adaptation of the Elo chess-rating system. Another is Sagarin's Predictor, a ranking based solely on point differential. (He also calls this White Owl or Rhein-gold because it's the column the cigar-chewing, beer-quaffing sharpies check out: Subtract the lesser team's figure from the better team's, adjust several points for the home court, and you have an instant line.) The third is the set of numbers that receives the most attention, his composite ratings, which weigh won-lost record and point differential equally. (The ratings are also published each Thursday in USA Today.)
Sagarin understands why a system based on points would give the gambling-averse NCAA the heebie-jeebies. That's why he proposes that his adaptation of Elo stand in for the RPI. Elo doesn't take into account margin of victory because, in chess, there is little significance to whether you win quickly or not; sometimes a player will sacrifice a queen to the larger objective of winning a game. Elo also recognizes that the higher-ranked team should win, so its ranking ought to rise less when it does. Conversely, the lower-ranked team should lose, so its ranking ought to fall less when it does. These brakes above and below give Sagarin's Elo a certain stability and guard against the stratification between the power leagues and everyone else that mars the RPI as the season progresses.
Most of all, Elo improves on the RPI by giving a win on the road more value than a victory at home—because home teams win 67% of the time. Over the 28 seasons Sagarin has analyzed the NCAA men's basketball rankings, the RPI has outperformed Elo only six times in forecasting how teams would do in postseason play. (Sagarin's composite fared even better, losing to the RPI four times; White Owl did best of all, losing just twice.)
"I respect what Jeff's doing," says Jim Sukup of The RPI Report. So apparently does the basketball committee, which since 1984 has requested his ratings for use during its deliberations. Sagarin says a member once told him that the committee likes his composite numbers precisely because they account for scores—"though," the committee member added, "we can't say that."
"If I were czar of the NCAA, I'd use Elo," Sagarin says. "It's not like I'm suggesting affirmative action for the small conferences. But the way the RPI is set up, it's like telling a kid in a ghetto high school, 'No matter what you get on your SATs, you can't go to college.' "