Merit aid is an increasingly important component of college scholarships, but policymakers are concerned that merit aid is often given to students who would enroll anyway. As a baseline we use a regression discontinuity (RD) framework to test an institution-level merit aid program at a public research university and find that the merit aid program successfully increases the likelihood of enrollment. We then add to the RD a structure that accounts for the probability that specific students would enroll (or not enroll) with certainty. This richer model, which allows us to identify students who are less certain about enrolling, indicates the merit aid is much more effective in convincing such students to enroll.