The accuracy of any particular procedure for predicting the outcomes of college football games can be evaluated empirically. In the case of a point predictor for the difference in score between
the “home” team and the “visiting” team, it is a simple matter of comparing the predicted differences with the actual differences. Among the criteria that might be adopted in the
empirical evaluation of the point predictor are the following three: (1) Percentage of winners = the percentage of games won by the teams “favored” to win by the prediction procedure; (2) Average absolute error = the average of the absolute values of the quantities obtained by subtracting the actual differences in score (between the home team and the visiting team) from the predicted differences; and (3) Root mean squared error = the square root of the average of the squared values of the quantities obtained by subtracting the actual differences in score from the predicted differences.
Among the various point predictors for the differences in score is that implicit in the betting line. The betting line is the “gold standard.” Its accuracy can serve as a basis for evaluating the accuracy of other point predictors.
Results on the accuracy of my college football predictions and (by way of comparison the accuracy of the betting line) over the course of each of 2 seasons (the 2018 and 2019 seasons) are included herein (under subheadings of the heading Accuracy). Starting in 2020 with the advent of the Covid-19 pandemic, college football has been subject to forces that have led to a great deal of instability. In addition to the pandemic itself, these forces include those directly or indirectly associated with conference realignment, with an acknowledgement of the players’ NIL (name, image, and likeness) rights, with (pandemic-related) extensions of players’ eligibility, and with the removal of any impediments that might have discouraged players from entering the so-called transfer portal. Such instability can add significantly to the challenges that one faces in predicting the outcomes of college football games. While the instability may have a negative effect on the accuracy of essentially all procedures for predicting the outcomes of college football games, it seems likely that model-based procedures would be among those affected the most.
Results on the accuracy of a predictor that are based on data generated during a period of instability would seem to be a relatively unreliable indicator of future accuracy. Results on the accuracy of a large number of procedures for predicting the outcomes of college football games over the course of the 2020 season and each of the subsequent seasons (procedures that include my procedure and the betting line) are among those found on the prediction tracker website (thepredictiontracker.com) maintained by Todd Beck. While these results may be of interest, they may be less indicative of the future accuracy of the procedures than would have been the case if the underlying data had been of a kind generated during a period of “lasting” stability.