It's the philosopher's stone of college football: the search for the perfect statistic to explain wins and losses, the ultimate wavering-quavering digit that could (aside from the score) point to broken part for the mechanically minded coach to replace and thus have the perfect win-manufacturing machine.
And like the philosopher's stone, that digit does not exist. Apart from the finally tally of points between teams playing in a single game, there's no single stat that explains it all, despite the efforts of a thousand statisticians to create one. Even the usual suspects frustrate expectations.
Like turnover margin, for example--if I give you the ball more often than you give me the ball, I should lose almost all of the time, no? Glen Mason, unemployed football coach, disagrees. Tops in turnover margin from the good end of the spectrum down from 2006:
2 Boston College
4 Brigham Young
6 Wake Forest
7 Western Mich.
10 Boise St.
A list of very good teams, yes. A list of top-tier monsters? Hardly, especially in the glaring cases of Minnesota, Syracuse, and Kentucky. We'd also like to note that coaching personality=stats here: there are some world class crotchety Matlock-watchers here, including Rich Brooks, Lloyd Carr, old-before-his-time Tom O'Brien, HOFer Chris Ault, and Jim Grobe. (As for Bronco Mendenhall and the bombs-away BYU offense...um, Mormon thrift, we guess?)
You actually have to do something with the ball in order to win. Even Methusalan geniuses like Ralph Friedgen, whose football education spans as diverse an ecosystem of gridiron thought as there is, can only come up with one stat that comes close: the 12 percent rule.
The statistic is derived by adding a team's interceptions, fumbles, dropped passes, sacks and penalties during a game and dividing that by the team's total number of offensive plays. The key is to keep the result under 12 percent -- meaning that the team is committing a human error on 12 percent or less of its plays.
It's Six Sigma for the shoulder pads set, but even then it's only hovering somewhere around a 90-95% effectiveness rate for predicting victory. When you have eleven variables interacting with eleven variables, each with a different task, route, assignment, and the ever-slippery element of human er-ROR involved, 90-95% may as close to holy as any grail-stat can be.
(Speaking of holy--holy hell, how in the fuck did Maryland win nine games last year?)