Suchergebnisse

We aimed to develop a framework for generating synthetic identifier datasets to support development and evaluation of data linkage methods. We evaluated whether replicating associations between attributes and identifiers improved the utility of the synthetic data for assessing linkage error.
We determined the steps required to generate synthetic identifiers that replicate the properties of real-world data collection. We generated synthetic versions of a large United Kingdom cohort study (the Avon Longitudinal Study of Parents and Children), according to the quality and completeness of identifiers recorded over several waves of the cohort. We evaluated the utility of the synthetic identifier data in terms of assessing linkage quality (false matches and missed matches).
Comparing data from two collection points in ALSPAC, we found within-person disagreement in identifiers (differences in recording due to both natural change and non-valid entries) in 18% of surnames and 12% of forenames. Rates of disagreement varied by maternal age and ethnic group. Synthetic data provided accurate estimates of linkage quality metrics compared with the original data (within 0.13-0.55% for missed matches and 0.00-0.04% for false matches). Incorporating associations between identifier errors and maternal age/ethnicity improved synthetic data utility.
Replicating dependencies between attribute values (e.g. ethnicity), values of identifiers (e.g. name), identifier disagreements (e.g. missing values, errors or changes over time), and their patterns and distribution structure enables generation of realistic synthetic data that can be used for robust evaluation of linkage methods.
Our framework provides a novel and generalisable mechanism for developing and benchmarking record linkage algorithms.

IntroductionCareful development and evaluation of data linkage methods is limited by researcher access to personal identifiers. One solution is to generate synthetic identifiers, which do not pose equivalent privacy concerns, but can form a 'gold-standard' linkage algorithm training dataset. Such data could help inform choices about appropriate linkage strategies in different settings.
ObjectivesWe aimed to develop and demonstrate a framework for generating synthetic identifier datasets to support development and evaluation of data linkage methods. We evaluated whether replicating associations between attributes and identifiers improved the utility of the synthetic data for assessing linkage error.
MethodsWe determined the steps required to generate synthetic identifiers that replicate the properties of real-world data collection. We then generated synthetic versions of a large UK cohort study (the Avon Longitudinal Study of Parents and Children; ALSPAC), according to the quality and completeness of identifiers recorded over several waves of the cohort. We evaluated the utility of the synthetic identifier data in terms of assessing linkage quality (false matches and missed matches).
ResultsComparing data from two collection points in ALSPAC, we found within-person disagreement in identifiers (differences in recording due to both natural change and non-valid entries) in 18% of surnames and 12% of forenames. Rates of disagreement varied by maternal age and ethnic group. Synthetic data provided accurate estimates of linkage quality metrics compared with the original data (within 0.13-0.55% for missed matches and 0.00-0.04% for false matches). Incorporating associations between identifier errors and maternal age/ethnicity improved synthetic data utility.
ConclusionsWe show that replicating dependencies between attribute values (e.g. ethnicity), values of identifiers (e.g. name), identifier disagreements (e.g. missing values, errors or changes over time), and their patterns and distribution structure enables generation of realistic synthetic data that can be used for robust evaluation of linkage methods.

Funded by ADR UK, a new data linking team at the Ministry of Justice set out to link administrative datasets across the justice space, for internal use and sharing with external researchers. To achieve this aim we sought a linkage implementation that was probabilistic, flexible, scalable and ideally open source.
Taking into account the tools available at the MoJ, existing open-source software (and paid alternatives) failed to meet our desired criteria. It was decided to develop a software package that builds on FastLink's implementation in R of an Expectation-Maximisation algorithm to estimate a Fellegi-Sunter linkage model, adding a range of technical improvements, increased functionality and customisation options. Distributed computing offered by Spark could facilitate comparable linkage jobs that run on much larger datasets and much faster. Working with government data, accountability and transparency are vital, so the data and models are made accessible by a range of intuitive visualizations.
The Splink python package has been downloaded 2 million times. This initially used Spark to deliver its superior performance, but Splink v3 caters for various SQL backends and more potential users. Splink and supplementary python libraries are publicly visible on GitHub and provide assistance in all aspects of data linkage:
- splink_data_standardisation - functions to perform general data standardisation
- splink_cluster_studio - creates an interactive HTML dashboard to analyse record clusters (building on splink_graph - a library for generating graph metrics)
- splink_synthetic_data - generating realistic synthetic data for testing linkage algorithms
- splink_demos - interactive demo/tutorial notebooks for a range of features of the Splink libraries
These tools are in continuous development and have already been used to deliver deduplicated and linked data products for the entire criminal justice system.
Through technical innovation and user-focused development, Splink has improved access to cutting-edge data linkage, and created groundbreaking research opportunities at MoJ and beyond. The team is grateful to ONS and other collaborators for testing and adopting these tools, and will continue to explore ways to improve performance and user experience.