I know some of the folks behind the KRACH are on this list, so I'm curious...
How well does the KRACH ratio predict win percentages when the disparity is large? Given the "Any Given Sunday" and "Goalie Stands on Head" scenarios which seem to happen somewhere in D1 hockey every weekend, it seems unlikely that we would (based on present ratings, not woofing) go 14-1 in a 15 game series with Clarkson, or 24-1 in 25 with SLU(t).
In theory KRACH should predict winning percentage exactly, but I've heard people on the USCHO board say that it hasn't done so well in past NCAA tournaments, though I don't have the stats myself.
Of course, something tells me RPI (the statistic) hasn't done that hot either :-).
I would think that at 11 games per year, it's a little tough to get a large enough sample to judge the predictive power of any rating system based solely on the NCAAs.
When you add up the several years together, you start to find a picture. And now we have 15 games per year :-D.