About four years ago, an engineer and a mathematician over at Georgia Tech began working on their own NCAA ranking system, called LRMC. They’ve since published a paper in an established research journal, proving that their ranking system is better than any of the other systems being used right now when it comes to predicting tournament play (RPI, or the AP poll, for example).
You can read about the system for yourself if you’d like, following the link above, but basically the LRMC system favors teams who win in blow outs more often than not; while overtime games are de-emphasized dramatically. (I’m sure this is a drastic over-simplification, but that’s all we have time for as I’m not prepared to deliver a lesson on Markov chains.)
But in this year’s first two rounds, the LRMC ranking system has taken some hits. Since the makers of LRMC want their rankings to be made available to the tournament selection committee, I think it’s fair to contrast the LRMC rankings to the tournament seedings. Of course, more often than not, the tournament seeding agrees with the LRMC ranking. But in the five first round occurrences where they disagreed, the higher seeded team won three times (UNLV over G.T., Kentucky over Villanova, and USC over Arkansas), whereas the team with the better LRMC ranking has only won twice (MSU over Marquette and Xavier over BYU).
Going 2 for 5 might not seem like a very significant loss for the LRMC, but the second round gave some very interesting results. This weekend featured three 5 seeds beating 4 seeds; thus the NCAA seeding committee went only one for four in these games. However, a 5 seed beating a 4 seed is hardly an upset, so the committee didn’t miss by much. The LRMC, on the other hand, got all four of these games wrong, VERY wrong. Here’s the four games, with the LRMC national ranking in parentheses:
- Butler (41) beat Maryland (10)
- USC (48) beat Texas (19)
- Southern Illinois (51) beat Virgina Tech (24)
- Tennessee (49) beat Virgina (33)
My point is that statistically speaking, the professors over at Georgia Tech can still claim their system is working on par with any other system, since when they are wrong, usually the tournament seeding is wrong too. But when you really look at the individual examples, when LRMC is off, LRMC is WAY off. Judging by those rankings, three of those four games should have been virtual locks. In fact, under the LRMC ranking system, at least two of our current Sweet Sixteen teams didn’t even deserve to be in the tournament field.
It pains me to say this, being an aspiring mathematician myself, but we still don’t have a formula that can fill out a bracket better than an average sports nut. At this point, it’s not clear that the LRMC’s bracket would beat the “cute mascots” bracket your secretary at work filled out.
But isn’t that what puts the “Mad” in “March Madness” anyways?