I say it's a ranking, but it's more of a projection... which is funny because the tournament projection is really more of a ranking. If I was to project the tourney field, I wouldn't just copy my tournament rankings, and if I was to rank teams mid-season I wouldn't just copy over my season projections. But people want season rankings and tournament projections, so they are called as such. Anyways, these team ratings are a blend of a few things...
-
Point Differential Adjusted for Opponent
-
Dominance Rankings
-
Shooting Luck Rankings
The first one - one might call it a simple rating system - is pretty standard. Point differential is a better predictor of the future than record, and, with the large disparity of talent in the college basketball, a strength of schedule adjustment is sorely needed. The latter two are explained below, and it just so happens their whole purpose is to account for things misconstrued by point differential (i.e. running up the score and hot shooting). Like most things on this site, the ratings are given in points above or below average per 100 possessions.
Each game has a point where the outcome is no longer in doubt. This can be at halftime when a team is up 50, or it could be as time expires with a game-winning shot. Teams that put away their opponents early are better teams. This point in time I define by a 98% win expectancy. The win expectancy is found by assuming a random walk for the remaining minutes. This point is then converted into a point spread value as not every minute is created equal (i.e. non-linear). For example, the difference between 0 minutes left and 1 minute left is much greater than 10 minutes left and 11 minutes left. Finally, this value is adjusted for opponent and tempo - and tempered if it was a really quick blow-out. The advantage of this system is that it doesn't assume all (insert margin of victory here) wins are the same; it ignores end-game occurrences like fouling and walk-ons.
Players don't always make open shots. Players also don't always miss contested shots. A free throw is an uncontested shot; you cannot defend it. And yet, sometimes the a team goes 10-20 from the line (and other times it goes 20-20). Sometimes you get lucky, other times not. There's variance. Now, 3-pointers are generally open shots and 2-pointers are contested a lot, but neither is only open shots or only contested shots. So, it would make sense to regress these statistics to take out the "luck" that has occurred. Six regressions are made - three types of shots for offense and defense. Regressions utilizes the attempts in the shooting statistic, the stabilization of the shooting statistic, the team's efficiency in the other shooting statistics, the opposition's efficiency in the shooting statistic, and (sometimes) blocks. You can then take the difference between the regressed shooting percentages and the actual shooting percentages and find how many points each team gained (or lost). This should make for more accurate projections. If you wrote Villanova off after this loss, maybe you wouldn't have knowing this statistic.