Title: It does what it says on the tin: safe synthetic data from coarsened margins

Abstract: This study introduces a novel approach to generating synthetic data (SD) that offers two distinct benefits to users over existing techniques. First, the method ensures transparency: recipients of the synthetic dataset are clearly informed about which variable relationships from the source data will be preserved with approximate accuracy. Second, it provides a safeguard that the synthetic data originates solely from information already assessed as safe regarding disclosure risk. The process begins by identifying and computing specific margins to maintain variable correlations within the SD. These margins then undergo statistical disclosure control (SDC) according to the data custodian’s protocols, which may include techniques such as top-coding, bottom-coding, merging small categories, and adjusting small counts. Additionally, it is recommended to further refine these curated margins by rounding all table counts to multiples of the disclosure limit. The Iterative Proportional Fitting (IPF) algorithm subsequently utilizes these adjusted margins to generate the synthetic data. The procedure is demonstrated through a practical example using data from the 1901 Census of Scotland.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC