Policy researchers need to be able to report the probability that their intervention "worked." For example, they might want to make a statement like: "There is an 83% chance that the effect of the intervention was greater than zero." We thought the p-value was a reasonable approximation of this kind of probability, but the ASA helped us understand that it's not. In order to calculate the probability that an intervention works, we need outside information about how often interventions work. In the Bayesian framework, this is called "prior information." But where does prior information appropriate for use in high-stakes evaluations come from? In this presentation, we will provide an example from the field of education policy to illustrate how to choose credible priors for policy research that are well grounded in quantifiable information - not personal belief.