Predictions for National Football League Games via Linear-Model Methodology

Abstract Results on mixed linear models were used to develop a procedure for predicting the outcomes of National Football League games. The predictions are based on the differences in score from past games. The underlying model for each difference in score takes into account the home-field advantage and the difference in the yearly characteristic performance levels of the two teams. Each team's yearly characteristic performance levels are assumed to follow a first-order autoregressive process. The predictions for 1,320 games played between 1971 and 1977 had an average absolute error of 10.68, compared with 10.49 for bookmaker predictions.