The Problem on Predicting the Outcomes of National Football League Games and Other Phenomena of Equivalent Interest

Attempts to successfully predict the point spread of National Football League games are viewed from the perspective of classical measurement theory. The hypothesis under investigation asserted that one factor limiting the valid predictions of the number of points scored by each team during games is the low levels of reliability inherent in the potential predictor variables. Data for all 28 teams of the NFL were collected for each of the 16 weeks of the regularly scheduled 1978-79 and 1979-80 seasons. Specific measures investigated included the number of points scored, the number of points yielded, the number of yards gained, and the number of yards yielded. These variables were then analyzed via Cronbach's alpha coefficient and other methods of assessing reliability. Results indicated that the a priori expectation of limited reliabilities was supported. The study discusses the implications of these results on traditional prediction strategies for both football specifically and psychological research in general, and offers an alternative approach to forecasting pro football games.