Data Linkages & Disclosure Risk

Abstract

Great varieties of data are now available to researchers. Analysts may wish to study outcomes in one dataset with predictors from another data source. Researchers have sought to combine data from many different sources. One potential linkage is to combine survey data with information from administrative sources. Linkages are now possible because data contain common identifiers. While linking data from different sources has great analytic potential, these linkages increase the risk of re-identification and disclosure. This paper discusses what aspects of data linkages increase disclosure risk. The paper examines how disclosure risk is related to the link variables themselves as well as the additional information that becomes available in the combined data. Even if the link variables are redacted from the combined dataset, disclosure risk may still be significantly higher because of the additional variables. The paper will also review different types of linkages including statistical matching and compare linking individual information and adding small area contextual variables. Illustrative examples are drawn from data deposited with the Data Sharing for Demographic Research (DSDR) project at ICPSR, University of Michigan. Furthermore, the paper includes suggestions about how to attenuate disclosure risk in linked datasets.