My March Madness

My graduate school was primarily a hockey school, although this year it has made the NCAA basketball Tournament for the first time in decades. In fact, my graduate institution plays my brother’s graduate institution in the first round (I don’t predict mine to win), and one of my grad school friends has his undergrad, graduate, and (present) faculty schools all in the tournament. (Take a look, though, at Chad Orzel’s bracket based on the strength of physics graduate programs: Cornell would win!)
While in grad school, I didn’t really follow the basketball team, and I don’t think I even went to a single game. But even though I don’t really, nor have I ever really followed college basketball, I will say that March Madness is the greatest sporting event in the world. Sixty-four games and they all matter.

My brother has, for a number of years, run the only March Madness pool that I participate in. Compared to the rest of my family at least (and sometimes his other friends as well) I tend to do rather well: I’ve never won but I have placed second twice. My general strategy is to pay as little attention to college basketball as possible during the regular season. This works: in our family, at least, there does seem to be an inverse correlation between the number of games watched and performance in the pool. Once, in graduate school, I tried to pick a bracket by flipping a coin, and it was absolutely dismal. I turn to two strategies, then, to fill out my brackets.

Statistics

There are two strategies with statistical bracket-picking methods. The first is to try find the characteristics (such as average winning margin or number of times the coach has been to the tourney) that historically have led to success in tournaments, and to see which of the current teams best meet the characteristics of historically successful teams. Pete Tiernan is the highest-profile guru of this sort of work and he’s put together a set of phenomenological models that predict success at all levels of the tournament, from choosing a final four to picking the 6-11 upsets. Of course, I’m too cheap to actually pay the $20 to buy full access to his research, nor do I want to buy ESPN insider to read his in-depth articles there. And the big question here is whether the methods actually work: compared with all the brackets on ESPN’s tourney challenge, a model (he has about a dozen) that hits the high 90’s pecentiles one year very often hits the 30 percentiles the next. So it may like picking winning lotto numbers: the winners don’t win because of the strength of the model, but because if there are enough entrants, one will be the best.

The second statistical approach is to construct models of team skill, and either rank the teams or put them head-to-head. An amazing amount of free analysis is available from Ken Pomeroy and I’m sure there are other sources as well.

For bracket construction, it’s actually pretty boring to just use somebody else’s ranking list to fill out a bracket. I’ve put together one bracket that’s a sort of half-hearted attempt to use Tiernan’s guidelines (at least the ones you can read for free) combined with Pomeroy’s Pythag numbers. What I discovered is that, more often than not, the simple guidelines don’t give clear-cut results, so doing this thoroughly requires sifting through an awful lot of data, which itself requires a good deal of effort to find. And although I’m a numerically-minded person, my interest does wane after a while.

Pundits

The method I like the most for bracket-picking is to see what all the sports pundits have to say. This year, at least, 5 writers for CNNSI and 5 writers for CBS Sports put their entire brackets up soon after Selection Sunday, and ESPN had 5 pundits with their Elite Eight picks. (CBS Sports has added two more brackets that I didn’t look at.) Sports pundits watch an awful lot of college basketball. To be a national-level sports pundit, you have to pay attention to all the conferences and a wide swath of teams. (This is where I think basketball enthusiasts stumble: they generally have their favorite teams and conferences upon which they focus their attention, and as a result overlook and underestimate the rest of the teams.)

What was interesting to me is the variation in pundit picks. All 5 of the CNNSI writers picked UCLA to win the tournament and none of the ESPN pundits did. All of the CNN pundits pick 13-seed Siena to upset Vanderbilt, while only 1 CBS pundit did. Four of five CNNSI pick 11th-seed St. Joseph’s to beat Oklahoma, while only one CBS did, while 3 CBS writers picked 11th-seeded Marquette to beat Kentucky, while only 1 CNNSI writer did. Out of all ten full brackets, there was only one prediction of a 14-over-3-seed upset: CBS writer Brian De Los Santos picked Georgia over Xavier.

In addition to sending them to my brother, I posted my brackets to the Washington Post Tourney Tracker: Search for thm_A_exp for the pundit-derived bracket, thm_C_stats for the statistics-derived bracket, and thm_B_pyth for the (boring to construct) bracket filled out strictly based on Pomeroy’s pythag statistic. I’m curious how each of these strategies fares in a wider pool of competition.