Hoops Nerd

What are the Hoops Nerd Ratings?

The Hoops Nerd Ratings are my college basketball rating system. They are expressed as either positive or negative points per 67 possessions. The goal is to measure how well a team can be expected to perform against average competition over the span of a 67 possession game, that being somewhere near the national average for tempo.

What makes the Hoops Nerd Ratings Unique?

The Hoops Nerd Ratings differ from other tempo-free efficiency ratings primarily in that current season data is transposed onto expected levels of performance and variance among the teams in each conference. That is, data from the past four seasons has been used to produce expected averages and standard deviations for each conference. By transposing current data onto these conference expectations in terms of quality and variance, we are able to properly situate each conference's average to the historical norm to which is it likely to regress over time, as well as get some insight into whether or not "breakout" teams are real.

For example, the Atlantic Ten has a very high expected standard deviation - because it is traditionally home to both very good (Xavier, for instance) and very bad (St. Bonaventure, for instance) teams - so if a team is having a real breakout season or a real stinker relative to its A-10 conference-mates, we can see that it is more likely to be maintained going forward than, say, a similar early-season outlier in the typically tight Big Sky Conference. The same principle applies to the transposition of the data onto the expected average of the conference.

This is essentially a contextual regression-to-the-mean process. We know how many points per 67 possessions that a z-score (number of standard deviations) is worth in each conference, and how well each conference is expected to perform, so Hoops Nerd Ratings take advantage of this.

How are the Numbers Made Early in the Season

The preseason ratings are made using a combination of the following factors:

• The previous season's data. The correlation of +/- points per 67 possessions for teams in the top 22 conferences is around .80 from season to season, so previous season's data is by far the most important factor;
• The incumbency rate of the given team's key contributors from last season, on a weighted-to-contribution basis, and
• The strength of each team's top two recruits.

Last season's data is adjusted based on incumbency rate and recruit rating, using historic year-to-year variance levels of teams within the given conference. The yields numbers that are transposed onto our expectations scale as described above, the same way that in-season data is. As the season progresses, less and less weight is attributed to the preseason ratings.

Daddy, Where does Raw Data Come From?

It comes from the lovely and talented Ken Pomeroy, whose adjusted efficiency statistics get squeezed, beaten and shaped through the methods listed above.