About This Blog

This is, truthfully, an exercise to keep me doing something "productive" when I feel like wasting time.  It is also an attempt to experiment with a rating system that intrigues me, but applied to other settings.  This is purely for personal amusement, and the Colley Matrix method is copyrighted by Wesley N. Colley.

The Colley Matrix method is a evaluative tool designed to better rate teams' relative strength using only win-loss records against individual opponents.  The Colley Matrix website applies this method to NCAA FBS football teams and NCAA Division I Men's Basketball teams.  Here, I have applied it to the 2010-11 NBA scores.  My spreadsheet can be found here, and you are welcome to copy it for your own manipulation.  If you are interested in updating the spreadsheet, or if you have a way to update the spreadsheet automatically, please let me know and I would be grateful.

The strength of the Colley Matrix method is that it provides a more accurate strength-of-schedule than a simple win-loss record.  The NBA is divided into two conferences of three divisions each.  From Wikipedia:

A team faces opponents in its own division four times a year (16 games), teams from the other two divisions in its conference either three or four times (36 games), and teams in the other conference twice apiece (30 games).

Thus, a team's win-loss record is affected significantly by its division's and conference's strength.  A good team in a weaker division will win more games than an equally good team in a strong division.

The Colley Matrix method is not a predictive instrument.  Studies have shown that point-differential is a much better indicator of future success, and others have utilized this with some success.  John Hollinger's ranking system (on ESPN) is not necessarily the best, but it is definitely the most accessible.  For more on what the Colley Matrix method does and does not do, please read Dr. Colley's thesis.

Intuitively, the Colley Matrix method produces better results given a larger sampling; early results are sketchy, and a team with a 7-0 record is much more likely to be ranked higher than a team with a 6-0 record, regardless of who they play.  Given the NBA's 82-game schedule, this should not be a problem.

I am curious to see how these results match up against an adjusted-average round-robin ranking of teams.  That is, if you took the average win-loss record of each team vs. each team--a 2-2 record becomes a 1-1 record, a 2-1 became a 1.33-0.66, and an 0-2 became a 0-2--how would those rankings compare to the Colley Matrix?  Both methods compensate for uneven pairings.  For this question, I could just use previous seasons, but I am not ready for that task quite yet.

I will blog about the results, posting the rankings and sometimes commenting on them as well.  I cannot promise a daily update, so copy my spreadsheet if you can't wait.

Peace, Chris

[EDIT:  In hopes of creating a better predicting instrument, I have designed the Colley-Woodrow Matrix method (often referred to as "Woodrow" or "CW" in tables).  Apologies for the name.  This matrix treats each point as a win or loss, so that teams routinely play about 200 "games" per game.  This allows point differential to matter while accounting for strength-of-schedule.  A team that scores +10 on a team that typically loses by -10 is not nearly as impressive as the team that scores +5 on a team that usually wins by +5.]