Abstract: This dissertation makes theoretical, methodological and empirical contributions to economic research by proposing a definition of influence as a reciprocal, heterogeneous force between individual agents. This formulation bridges the gap between the sociological meaning of the term and the economic concept of an externality. In my theoretical contribution, I use an evolutionary game theory framework to show that under weak and heterogeneous influence coordination can emerge in the aggregate even if two-player games are not coordination games. This has both economic and social consequences. While players can avoid the coordination problem of inefficient local Nash equilibria they cannot solve the externalities problem of optimizing private rather than social surplus. Under heterogeneous influence it is also possible that players choose strategies against their type in equilibrium. I make use of the theoretical findings to approach the empirical task of separating influence from individual preference. Using U.S. Senate roll call votes I calculate an individual Partisanship Score for each senator and show that senators reduce their level of partisanship over the course of a six-year term if they stand for reelection, meaning legislators either move from voting their own preference or their party preference to voting the constituency preference. The instrument to separate these two explanations is the voting record of retiring senators. Absent outside influence retiring senators can vote their own preferences, and I observe that retiring Republicans vote with their non-retiring peers, indicating that they are subject to party influence at the beginning of each term. No comparable pattern is detected for retiring Democrats. The empirical findings depend on the ability of the Partisanship Score to detect cyclical changes in the voting behavior, so I test the score against the two most common ideological scores, ADA and W-Nominate, using a Monte Carlo simulation. I find that the Partisanship Score is better able to detect deviations from voter preferences if votes are not easily projected onto the unidimensional liberal-conservative policy line, or when caucus majorities are large.