ABSTRACT
The Minnesota Population Center (MPC) has released linked datasets through its NAPP and IPUMS projects, making them readily accessible to researchers. Prior to the availability of complete count census microdata from the MPC, researchers applied various forms of record-linking software. This essay describes the techniques used in the MPC's linking program and briefly compares this technique with those used by other researchers. The key feature of the MPC linking method is the construction of cumulative name similarity scores, based on approximately 2.5 billion record comparisons; we also use support vector mechanics to classify potential links. This article explains modifications made for the final linked datasets and includes a discussion of the role of weighting variables when using linked data.
