An explanation of leadership within a particular class of institutions is developed. The institutions of interest are non-hierarchical, endow their leaders with few formal powers, and are hampered by principal-agent problems. The explanation proceeds in two stages. First, game-theoretic models of leaders and followers engaged in two social dilemmas are developed. The models define leaders as the set of actors who can send cheap-talk signals to a group of followers. Leadership occurs when followers, recognizing that their chances of success without a leader are slim, choose to voluntarily accept a leader's advice. The models highlight the importance of the underlying social dilemma, either a coordination or collective action problem, as well as the leader's reputation for motive and competence. Second, hypotheses are derived from the models and are tested under laboratory and field conditions. The experimental results demonstrate that leaders are remarkably adept at solving simple coordination problems, provided that followers have sufficient confidence in the leaders' reputation for motive and competence. The effectiveness of leadership decreases, however, in collective action problems. In these settings, followers are often unwilling to take the risk that is inherent in following a leader's advice. Roll call data from the U.S. Senate are used to check if the experimental findings extend to a real-life setting. The data suggest that the models and laboratory evidence possess considerable levels of external validity.