Incorporating Record Linkage Measurement Error into Descriptive Network Metrics (306448)*Abby Smith, .
Keywords: record linkage, misclassification, social statistics
Record linkage, also known as entity matching, is imperative for many applications, ranging from Census work to the creation of health indicators and involves finding recovering unique individuals from multiple datasets. Network analysis has become increasingly relevant; we are no longer only interested in properties of individuals, but their connections as well. Record linkage is susceptible to measurement error. Letters can be out of order or missing, punctuation can be misplaced, and abbreviations can be varied. Record linkage error is typically reported in terms of precision/recall as a ‘F1 measure’, but little work has been done to assess how these standard reported metrics of linkage quality can inform network analysis, such as the determination of structural and node metrics. Additionally, we seek to develop “best reporting practices” for linkage error for users who hope to conduct network analysis.