Causal inference in observational studies is notoriously difficult without physical randomization of treatment. Yet when the treatment of interest cannot be randomized, as is often the case in political science, randomized instruments may enable scholars to draw causal inferences of theoretical interest. We illustrate a Bayesian framework developed in the statistical literature of analyzing, making explicit, and relaxing crucial assumptions that are implicit in virtually all instrumental variables analyses. We provide an application to assess the causal effect of racial perceptions on answers to a political knowledge survey. The application demonstrates the usefulness of this framework in permitting researchers to gain more leverage over theoretical quantities of interest with greater scientific credibility.