Frustrated that you didn't do as well in your March Madness office pool as you'd hoped? Next year you might want to try making your picks using a computerized ranking system built by professors at Georgia Tech.
The system, called LRMC for the two mathematical processes it uses (logistical regression and Markov chain), correctly picked all four of the semifinalists in this year's National College Athletic Association (NCAA) tournament, teams colloquially known as the "Final Four." Moreover, it has predicted semifinalists with 83 percent accuracy over the past nine years, according to its creators.
Out of the 36 Final Four teams in the past nine years, the system has accurately picked 30 -- results that are more accurate than the AP poll of sportswriters and the ESPN/USA Today poll of coaches, the Ratings Percentage Index (RPI) formula and the tournament seeding system, said Joel Sokol, a professor of industrial systems and engineering at Georgia Tech and one of LRMC's creators. Over the same time frame, the seedings and polls have picked 23 of the final-four participants, and the RPI identified 21, Sokol said.
In addition to accurately predicting the Final Four this year, the system also is one game ahead of the seeding system and three games ahead of the RPI, he said. Moreover, in games when the LRMC system and the polls and seedings disagreed, the Georgia Tech system chose three out of five correct against the seeds and five out of seven correct against the RPI, Sokol added.
In the U.S., the annual NCAA basketball championship, also known as March Madness, pits 64 teams against each other in a complex bracketing and seeding system based on the teams' performances in the regular college-basketball season. The tournament is popular fodder for office betting pools, even for people who are not necessarily interested in sports.
Sokol said he and another professor, Paul Kvam, came up with the idea of the ranking system several years ago, when Georgia Tech's basketball team was not considered for the NCAA tournament. At the time, some pundits said if the team had won one more game, they would have been a "bubble" team, the term for teams that are in the running for a March Madness slot.
However, to Sokol and Kvam this seemed unfair, because "there was a game where Georgia Tech was playing Tennessee and Tennessee won the game on a last-second, half-court shot."
"That really improbable shot made the difference, and [we wondered] if that really indicates that Georgia Tech is somehow a worse team than if that guy had missed the shot," Sokol said.
The two devised the system and began using it in the 2003-2004 college basketball season; eventually another professor, George Nemhauser, joined the LRMC team.
About halfway into the college basketball season in January, the LRMC begins collecting information from the NCAA scoreboard on Yahoo. The logistical regression part of the system analyzes information about the teams, the players, where the game is played, which team has home-court advantage, what the margin of victory in the game was and the like, Sokol said.
"Based on that information, it estimates what the probability is that the team that won is actually better or ... if the team that's worse just happened to win," he said. The part of the system that uses the Markov chain process then takes those estimates and puts them together into a ranking, Sokol added.
Once the system was put into use, Sokol and his team went back and applied it to NCAA tournaments back to the 1999-2000 college basketball season to track the accuracy of results over the years. He said also that once the LRMC was launched, he and his colleagues stopped playing NCAA tournament betting polls, believing they have an unfair advantage over other players.