Hey James! So this model is separate from ELO. ELO is solely a measure of an individual team’s strength. This model uses several metrics, both for the teams themselves and how a team compares to their opponent. For example, each team’s Defensive PPA (predicted points added), the current point-spread, the difference in the two teams’ FPI ratings and SRS ratings (from College Football Reference) are incorporated, along with a few other metrics. These data then get fed into a linear regression algorithm to predict a point spread.
Think about it like this - a team may be an all around “weaker” team, but based on certain characteristics may be able to exploit the weaknesses of its stronger opponents. This also takes into account home field advantage, rest after a bye week, etc.
Elo can be viewed as an overall power rating, but this model is better at judging two teams against each other based on various team and situational factors.
Something seems off with the Colorado @ Texas Tech game. Colorado has a ~40 point advantage in ELO but they are 22.6 point underdogs?
Hey James! So this model is separate from ELO. ELO is solely a measure of an individual team’s strength. This model uses several metrics, both for the teams themselves and how a team compares to their opponent. For example, each team’s Defensive PPA (predicted points added), the current point-spread, the difference in the two teams’ FPI ratings and SRS ratings (from College Football Reference) are incorporated, along with a few other metrics. These data then get fed into a linear regression algorithm to predict a point spread.
Think about it like this - a team may be an all around “weaker” team, but based on certain characteristics may be able to exploit the weaknesses of its stronger opponents. This also takes into account home field advantage, rest after a bye week, etc.
Elo can be viewed as an overall power rating, but this model is better at judging two teams against each other based on various team and situational factors.
Thanks for your interest!