An assessment of path dependence in collective decisions: evidence from football polls

Abstract

While the topic of path dependence has reappeared in some macroeconomic circles, the existence of path dependence of voting outcomes is still more a matter of speculation than of evidence in spite of its importance to Arrow's Impossibility Theorem. This analysis considers the empirical relevance of path dependence, drawing data from Associated Press college football polls. In major college football, the primary measure of success is a team's year-end ranking by sports writers and/or coaches, ostensibly based on performance. This article examines the degree to which the year-end orderings for teams, R ik, are influenced by initial, pre-season rankings, Ri 1, and mid-season rankings, Rim, using these 'Top 20' votes for 1980 - 1989, while accounting for various measures of team performance. In addition to OLS estimates, cumulative logit models (CLM) and censored logistic regression (CLR) are used to deal with ordinality and sample problems. The results in Section III indicate a relationship between pre-and post-season votes strong enough to have influenced many polls' outcomes, including the designation of the 'national champion' in at least one year of the sample.