College basketball rankings: Introducing T-Rank

One of Ken Pomeroy's interesting discoveries is that the preseason AP poll is a really good indicator of team strength -- maybe even better than the end-of-season AP poll. At first blush, this seems crazy. How can our opinions about team strength get worse after spending six months watching thousands of games?

Well, the hypothesis is that the preseason ratings are based on more objective, reliable data. Things like traditional strength of program, quality of recruits and returning players, returning minutes, etc. Once the season starts, however, our opinions of team strength become tainted by randomness: the results of actual games. These results overwhelm our objectivity, inducing us to create narratives about "heart" and "knowing how to win." But most of these narratives are just clever arrangements of statistical noise, like faces in things.

The upshot is that there is a place for more data-driven, formulaic preseason ratings. The AP preseason poll is good because human beings are very good at the kind of fuzzy logic required to take in a bunch of disparate data and form a judgment, so it is worth paying attention to. But computer ratings like Pomeroy's, Dan Hanner's and Team Rankings -- each of which use different models -- can provide different perspectives, and maybe some surprising insights in the process.

On a "the more the merrier" principle, I created my own spreadsheet-driven preseason ratings, which I call the T-Rank (T is for "terrific," of course). The methodology takes a variety of data (stats for every returning D-1 player, team returning minutes, incoming recruits, transfers in and out, weighted historical program performance, program momentum and loss of exceptional players) and spits out KenPom-style efficiency projections and a pythagorean winning percentage for every team. T-Rank is undoubtedly inferior to Pomeroy's and Hanner's preseason ratings, but the important thing is that it's a lot of fun -- and it gives me plenty of fodder for writing preseason college basketball posts.

The full 351-team T-Rank is available here, but I'm going to start with some analysis of our favorite league, the Big Ten.

Thankfully, Wisconsin comes out on top. But then things get pretty interesting. The recent strength of Michigan State, Michigan and Ohio State powers them to top-20 ranks overall, despite fairly significant player turnover. Then Minnesota, Iowa, Nebraska and Indiana fill out the rest of the league's projected tourney teams. Nebraska is lower here than in most human projections, and this is frankly because it's hard for T-Rank to capture an abrupt turnout around like Nebraska's and project it into the future. But the Huskers did actually finish No. 44 in the final Kenpom ratings last year, so T-Rank projects them to improve a little.

Using the T-Rank efficiency projections, I can run simulations of the entire Big Ten season. Here are the results in terms of expected wins:

Here again we see a fairly clear distinction between the top four of the league and a muddled middle. Despite having essentially the same T-Rank as Indiana, Nebraska is projected to win one less game because it has easily the toughest schedule of any of the good teams in the Big Ten.

This is another X-Factor to keep in mind this year. With the introduction of Maryland and Rutgers to the league, the schedule can get quite unbalanced. Assuming the T-Rank is accurate, here are the projected strength of schedule ratings (red is hard, green is easy):

Wisconsin, of course, benefits because it doesn't have to play itself, but it also has by far the easiest slate of games among the contenders. Northwestern, which T-Rank projects to be bad, also suffers from the most difficult conference schedule.

Now, I know what you're thinking: with the invention of T-Rank, what is the point of actually playing the games? This is a question for the philosophers to ponder, not me.