Following a discussion about the RPI ratings, i looked
into the final season ratings for and calculated the
standard deviation for the various %'s. The standard
deviation is a quantitative measure of the variation in
a data set.
Win% OppWin% OppOppWin%
std. dev. (in %) 17 3 1
One can thus say that Win% is 50%+/-17%. The more
variables one averages over, the smaller the deviation
will be and this explains why for the strength of schedule
variables the deviation is much smaller.
RPI gives different weights to the different categories.
If we assume that the effective weight is the weight
corrected by std. dev. ( i believe that this assumption is
justified) we find
weight Win% OppWin% OppOppWin%
RPI 25 50 25
Effective 71 25 4
RPI thus gives Win% a much larger weight than might be
thought by looking at the 25% weight.
Eli