Right now in my rankings, Florida, Xavier, and Connecticut are running the show, and their success has also had the misfortune of elevating some teams lucky enough to secure a victory over any of these three thus far in the season. Take a look at some of the rankings I have cherry picked.
[table “7” not found /]This is the result of the clustering problem in action. As I described some on my method page, early in the season, my rankings may consider some groups of teams to be particularly strong because they have performed roughly equally against each other, and in their limited play against opponents outside the cluster, they have been very successful. This is what has happened with Florida, Xavier, and Connecticut thus far in the season.
The triangle in the middle is the cluster I am talking about. The direction of the arrows, here, represents a win. Notice how Connecticut has beaten Florida, Florida has beaten Xavier, and Xavier has beaten Connecticut. Though the calculation takes into account more than a binary win or loss, roughly speaking, these teams are equal to the algorithm!
So the question is now, How does this cluster compare to the outside world? Well, the majority of these teams’ losses come from within the cluster. These teams have defeated almost all outside opponents aside from a few losses, only some of which are bad (Xavier to Green Bay, Connecticut to Iona, Connecticut to Saint Joseph’s). So the result is that the algorithm assumes that these three equal teams must be a very successful group, and there they go to the top ten of the rankings.
Notice a second triangle between Florida, Saint Joseph’s, and Connecticut. Saint Joseph’s is a terrible team (they have two wins!!!), but they are ranked 56 in my most recent rankings. There is not even a team with three wins ranked above Saint Joseph’s. And this is all because they were fortunate enough to get entangled in another one of these triangle of success with two very highly rated teams. So the algorithm assumes that Saint Joseph’s must be one on par with these two top-ten teams, and this entanglement is enough to offset their seven losses in nine games to nearly place them in the top 50 in my rankings.
Florida State has also stood to benefit from this cluster, but perhaps they have more merit. They have a played a difficult schedule thus far, and have recorded a strong win over Purdue and a difficult road win at Florida, but also an unseemly loss against Indiana this week. They are certainly a good team, but that road win at Florida is probably weighted disproportionately in their rating due to this here cluster.
I’m secretly hoping one of these teams suffers a big loss in the near future to recalibrate my rankings. Florida has a tough few games coming up, most notably away against Butler, and UConn will have to face Indiana next week. Perhaps that is not in the spirit of trying to create an objective ranking system. But who knows, maybe one of these teams will surprise college basketball fans as the season progresses. This early in the season, and particularly with this cluster, however, I would not place too much faith in my model.