The model’s off to a strong start, and we introduce schedule adjustments this week
Of course, every model produces simultaneous reactions of “How could you have Team X so low” and “How could you have team Team Y so high.” The most fun version of this is when Team X beat Team Y. Two cases in point: Clemson still leads UGA in the model, and GT is well ahead of NIU.
What’s going on? A model like this one is not a power ranking based off of past results. It is meant to be forward-facing and predictive. In the case of GT and NIU, NIU won by one point in a game that the underlying statistics show that GT would win 75% of the time. The model reflects that, as the current ratings would still have GT as an 11 point favorite over NIU on a neutral field. And that should track with our intuition that if GT played NIU again, they would still be a significant favorite.
In the case of UGA and Clemson, UGA has clearly performed better so far this season. Because of that, they have made up about 4 points on Clemson compared to where they were in the preseason. The preseason ratings still hold about a 70% after just two weeks of games to avoid overreacting to a singular positive or negative performance. If Georgia and Clemson maintain their courses of play for the rest of the season, Georgia will of course pass Clemson. But as of now, the model would still give Clemson a slight edge on a neutral field.
Of course, no model is perfect, and in the first season of running mine, I know that it has significant weaknesses. But we can track the model’s performance by analyzing how it does against the Vegas spread over the course of the season, and how it does compared compared to the actual point margin of each game. Winning 55% of games ATS is typically considered the gold standard of performance, and hitting an absolute error of 12.5 points per game is excellent.
The college prediction tracker keeps up with ATS and absolute error performance of almost 50 different models. After two weeks, the Binion Index would be 2nd overall ATS and above average in absolute error.
Season to Date ATS: 50-41-5 (54.7%)
ATS Goal: 55%
Season to Date Absolute Error: 13.95
Absolute Error Goal: 12.5
- There’s some bigger changes this week, as we introduce an adjustment for strength of schedule and continue cranking up the weighting giving to current season games in comparison to preseason ratings.
- There are still a number of games with no play by play data (including 2 teams who don’t have data for either game), so a few teams (i.e. San Jose State) are way off where they would be if the data was available.