<p>I have a vague idea of how one might do the 5%X, 15%Y voodoo … but have never seen any documentation on it. I suspect they start by identifying a subset of schools they think are “best”. They then hypothesize a first-pass set of features (class size, salaries, etc.) they think measurably describe the goodness of those schools. They experiment with many combinations of weightings until they arrive at one that appears to generate a plausible ranking, one that roughly corresponds to their concept of Ground Truth. They add/subtract features (graduation rates, etc) and re-adjust weightings, in a long game of whack-a-mole, until they arrive at a stable set of features and weights (one that continues to output a plausible ranking as you introduce new schools into the input set). Something like that … but maybe somebody here has better insights into the process.</p>