Nonresponse error is a common but undesirable feature of a survey. Survey practitioners use various techniques to reduce non-response error. In the National Compensation Survey (NCS), the state and local government frame is developed using administrative files maintained by the states for unemployment insurance (UI). In this paper, we explore the extent to which a frame refinement process, used to reconfigure UI reporting units so that they conform to how payroll records are kept in an establishment, improved the ability to collect data for the establishment as sampled, thereby improving response as well as reducing respondent burden and the need for post-collection sample-weight adjustment. An overview of the refinement process, the system tool developed for refinement, and several approaches to measuring the impact of frame refinement on non-response will be explored.