Post by gogophers on Feb 26, 2017 20:26:45 GMT -5
This weekend's edition of the Wall Street Journal reports that the NCAA is considering getting rid of RPI for seeding the men's basketball team, perhaps as early as next year. It's too soon to say RPI, RIP, but NCAA does seem to be wondering aloud, at least in a context that it actually cares about, March Madness, whether there's a better measure. Most posters on this board think they know the answer. Some quotes from the article, with some familiar refrains:
"As another edition of March Madness looms, the NCAA is accelerating the search for new cutting-edge data tools to better measure which teams should be invited to the annual college-basketball extravaganza and which ones should stay home.
When the selection committee convenes next month to pick teams for this year’s tournament, it will again lean heavily on a 36-year-old formula called Rating Percentage Index. Used since 1981, it attempts to take 351 Division I teams that play vastly different schedules and rank them based on their performance and the quality of their opponents.
Despite numerous tweaks over time, it has faced frequent criticism for being outdated in today’s world of rapidly advancing sports analysis. Unlike more sophisticated metrics, RPI only accounts for whether a team wins, not the margin of victory, and is widely considered to use an overly simplistic method for determining schedule strength.
Changes to this archaic system could be in place as early as next March. To devise alternatives, the NCAA last month brought together some of the most prominent analysts and mathematicians in the college basketball world.
....
There’s no doubt that the different ranking systems can have varied results with real-world implications. The RPI, for instance, ranks West Virginia as the 30th-best team in the country, while Pomeroy has the Mountaineers at No. 3. A composite metric would compensate for such a wide discrepancy.
Before that can happen, the NCAA first must answer a basic question: What is the purpose of this new metric?
“They have to understand what they’re trying to pick for the tournament,” Pomeroy said. “Are you trying to pick the teams that are most accomplished? Or the best teams the moment the tournament starts?”
This is a subtle but significant distinction. The RPI and KPI are results-oriented metrics, which means they reward and penalize teams based on the outcomes of their games, the quality of their opponents and where the games are played. KenPom and Sagarin are predictive metrics, which means they also factor in the scores of games.
The difference explains the disagreement over West Virginia: Results-oriented metrics see a team with six losses, while predictive metrics see that the Mountaineers lost those six games by an average of 3.7 points.
“Having multiple data points in a conversation allows you to minimize and identify outliers,” Pauga said. “The RPI is quantifying results, whereas a metric like KenPom is ranking the quality of teams, and those two don’t necessarily coincide.”
. . . .
In the meeting, the attendees suggested the possibility of devising two composites, one based on results-oriented metrics for picking teams and the other on predictive metrics for seeding them within the bracket. That idea was met with concern that it would only add to the confusion. But it wasn’t ruled out.
. . . .
When it deliberates, the selection committee currently refers to team sheets that show, among other things, a team’s record against opponents ranked in the top 50 and bottom 200 of the RPI. That figure has historically been important in determining who winds up reaching the tournament, which the statisticians say is a problem: RPI is relatively easy for teams to manipulate with strategic scheduling.
Moving forward, these team sheets could be built not around RPI, but a composite metric, which would go a long way toward improving the tournament selection process by more precisely revealing “good wins” and “bad losses.”
"As another edition of March Madness looms, the NCAA is accelerating the search for new cutting-edge data tools to better measure which teams should be invited to the annual college-basketball extravaganza and which ones should stay home.
When the selection committee convenes next month to pick teams for this year’s tournament, it will again lean heavily on a 36-year-old formula called Rating Percentage Index. Used since 1981, it attempts to take 351 Division I teams that play vastly different schedules and rank them based on their performance and the quality of their opponents.
Despite numerous tweaks over time, it has faced frequent criticism for being outdated in today’s world of rapidly advancing sports analysis. Unlike more sophisticated metrics, RPI only accounts for whether a team wins, not the margin of victory, and is widely considered to use an overly simplistic method for determining schedule strength.
Changes to this archaic system could be in place as early as next March. To devise alternatives, the NCAA last month brought together some of the most prominent analysts and mathematicians in the college basketball world.
....
There’s no doubt that the different ranking systems can have varied results with real-world implications. The RPI, for instance, ranks West Virginia as the 30th-best team in the country, while Pomeroy has the Mountaineers at No. 3. A composite metric would compensate for such a wide discrepancy.
Before that can happen, the NCAA first must answer a basic question: What is the purpose of this new metric?
“They have to understand what they’re trying to pick for the tournament,” Pomeroy said. “Are you trying to pick the teams that are most accomplished? Or the best teams the moment the tournament starts?”
This is a subtle but significant distinction. The RPI and KPI are results-oriented metrics, which means they reward and penalize teams based on the outcomes of their games, the quality of their opponents and where the games are played. KenPom and Sagarin are predictive metrics, which means they also factor in the scores of games.
The difference explains the disagreement over West Virginia: Results-oriented metrics see a team with six losses, while predictive metrics see that the Mountaineers lost those six games by an average of 3.7 points.
“Having multiple data points in a conversation allows you to minimize and identify outliers,” Pauga said. “The RPI is quantifying results, whereas a metric like KenPom is ranking the quality of teams, and those two don’t necessarily coincide.”
. . . .
In the meeting, the attendees suggested the possibility of devising two composites, one based on results-oriented metrics for picking teams and the other on predictive metrics for seeding them within the bracket. That idea was met with concern that it would only add to the confusion. But it wasn’t ruled out.
. . . .
When it deliberates, the selection committee currently refers to team sheets that show, among other things, a team’s record against opponents ranked in the top 50 and bottom 200 of the RPI. That figure has historically been important in determining who winds up reaching the tournament, which the statisticians say is a problem: RPI is relatively easy for teams to manipulate with strategic scheduling.
Moving forward, these team sheets could be built not around RPI, but a composite metric, which would go a long way toward improving the tournament selection process by more precisely revealing “good wins” and “bad losses.”