The forecasting accuracy and determinants of football rankings

Numerous models vie to explain the extent to which and the manner in which people use new information to reconsider existing beliefs. We analyze week- to-week changes in the rankings of big time college football teams to test predictions based on these models. We test logistic regression models of wh ether Top 25 teams moved up a certain number of places in the rankings foll owing victories, using data on weekly movement in the AP rankings, 1985-95. The predictors in these models are indicators of inertia and constraint, r ank elevators, and the passage of time. Lower-ranked teams move up faster a fter a victory than do higher-ranked teams, but moving up in the rankings a fter a victory is an incremental process - much more incremental than movin g down after a loss. All the focal predictors in the models perform as expe cted in influencing a team's chances of moving up following a win. That is, the odds of moving up are greater: the fewer prior losses a team has suffe red; the lower the team's rank before the victory; if an opening has occurr ed higher in the rankings; if the victory is over a ranked opponent, and es pecially a higher-ranked opponent; and if the standing of opponents played earlier in the season has risen. These results are most compatible with a m odel that combines what are often treated as contradictory ideas - that peo ple are 'naive scientists' or 'intuitive statisticians', on the one hand, a nd that they are extremely conservative information processors, on the othe r. (C) 2001 International Institute of Forecasters. Published by Elsevier S cience B.V. All rights reserved.