Thursday, October 15, 2009

Degrees of Separation

Ratings systems are complicated beasts, but one thing that is certain is that for a system to be accurate in rating the teams and predicting games, there has to be enough data.  But more importantly, when comparing two specific teams with a pool of teams as large as the NCAA has, you need to have a linkage between the teams and ideally there are enough linkages that are short enough for the comparison to be meaningful.

Something I track to help with this is the degrees of separation between teams playing each week.  The term "degrees of separation" was popularized a number of years ago with the Six Degrees of Kevin Bacon trivia game.  And what I do with predictions is basically the same thing.  Let's take a look at an example.

In tonight's game between Cincinnati and South Florida the computer is picking South Florida by 0.8 while the line is Cincinnati by 3.5.  That much of a difference seems pretty significant, but how sure should we be of this pick?  How many degrees of separation are there between the teams?

A quick check reveals that the shortest path is 4 and there are three ways to accomplish it:
  • South_Florida-->Wofford-->Wisconsin-->Fresno_St-->Cincinnati
  • South_Florida-->Western_Kentucky-->Florida_Int'l-->Rutgers-->Cincinnati
  • South_Florida-->Syracuse-->Northwestern-->Miami_OH-->Cincinnati
There are also 9 paths with 5 degrees.

What does all this mean?  Well, 4 degrees of separation is still quite a bit and having 4 paths that are pretty distinct helps but ideally there are fewer degrees of separation so my computer would put a moderate probability on the pick being correct.

The degrees of separation, along with other factors including the predicted margin, the Vegas line, and how predictable the teams are, all enter in to the best picks and locks I publish each week in my newsletter.  Take a look and subscribe if you are interested.