Current models for predicting US presidential elections focus on economic growth, the popularity of the current President, and incumbency. Most presume implicitly that the election is a referendum on the performance of the sitting President and his party. Characteristics of the opponent and personal characteristics of the incumbent party's candidate are not accounted for. In this paper, we explain and test a new model drawn from the literature on the performance of corporate executives. The model relies on objective measures of the complexity with which each candidate processes information. In all presidential elections for which these measures are available, the candidate who demonstrated the greater complexity of information processing won. Adding the information processing assessment also improves the predictive performance of current models. These findings suggest that the referendum model is needlessly limited, and that voters in fact compare the expected performance of both candidates before making a decision.