I'm learning/reviewing MCMC methods now for work and...how about a MCMC algorithm that takes in basketball box scores and attempts to generate estimates of each player's ability?

You'd probably have to deal with too many parameters (one for each player in the league)...or maybe not?

Pardon the vagueness, I'm only suggesting this as a theoretical exercise.

EDIT: well, that didn't take long. http://www-stat.wharton.upenn.edu/~zhangk/2010%20Scoring%20a...