CFB Model by Dunham

The core tenet of the model is to devise a power rating system based on what has changed since the end of the previous season. As a power system, past wins and losses do not matter, with predicted wins and losses coming as an organic output of the power and the schedule. Only FBS teams are explicitly rated, though FBS vs FCS games are included in the modeling.

Data for every metric is converted to a common scale (last season’s final power ratings) in order to quantify, compare, and combine. Duplication caused by overlapping metrics (such as Returning Starts and Seniors) is removed or minimized.

Each metric is weighted by using a Pearson correlation coefficient. A good amount of history is gathered for each metric, and its correlation to a composite power rating (Sagarin + Massey) is computed across those years. Four stripes of correlation are computed:
        Value of the metric vs Value of the power rating (Value-Value)
        Value of the metric vs Y-over-Y change of the power rating (Value-Delta)
        Y-over-Y change of the metric vs Value of the power rating (Delta-Value)
        Y-over-Y change of the metric vs Y-over-Y change of the power rating (Delta-Delta)
The highest of the four coefficients is used. The stripe of correlation also determines how the metric is converted to the common scale (last season’s final power ratings).

Metrics for 2025