Hausdorff Distance-Based Record Linkage for Improved Matching of Households and Individuals in Different Databases

Matching households and individuals across different databases poses challenges due to the lack of unique identifiers, typographical errors, and changes in attributes over time. Record linkage tools play a crucial role in overcoming these difficulties. This paper presents a multi-step record linkage procedure that incorporates household information to enhance the entity-matching process across multiple databases. Our approach utilizes the Hausdorff distance to estimate the probability of a match between households in multiple files. Subsequently, probabilities of matching individuals within these households are computed using a logistic regression model based on attribute-level distances. These estimated probabilities are then employed in a linear programming optimization framework to infer one-to-one matches between individuals. To assess the efficacy of our method, we apply it to link data from the Italian Survey of Household Income and Wealth across different years. Through internal and external validation procedures, the proposed method is shown to provide a significant enhancement in the quality of the individual matching process, thanks to the incorporation of household information. A comparison with a standard record linkage approach based on direct matching of individuals, which neglects household information, underscores the advantages of accounting for such information.
Comment: 20 pages, 5 figures, 10 tables