Power law distributions suggest that if you can get a few datapoints much of the rest is roughly estimable.

Rankings are less utterly useless than first appears.