It is widely recognised that greater publication, sharing and linking of existing data sources holds considerable potential to increase transparency, improve service delivery, transform policy outcomes and help to drive innovation, productivity and economic growth.
Subject to strict safeguards and where there is significant public benefit, the Australian Bureau of Statistics (ABS) is increasingly making use of record linkage techniques to combine existing sources of data, for the purpose of producing analytical datasets that have enhanced temporal and cross-sectional detail. Frequently this record linkage must be achieved without the benefit of unique or definitive linkage keys, and consequently incorrect links may result. The proportion of links that are correct, or the ‘precision’ of the record linkage, can be difficult to establish when even careful clerical review may fail to resolve whether or not links are correct. Measures of precision are useful for deciding whether to proceed with a record linkage project, for evaluating alternative linking strategies and for establishing quality measures for estimates based on the linked data. This paper proposes an estimator of precision for a linked dataset that has been created by either deterministic (rules-based) or probabilistic record linkage. Both methods are widely used at the ABS. The paper shows that the proposed estimators perform well in simulation, and it is envisaged that the proposed estimator will be part of the ABS’ record inkage tool kit.
Follow us on...Like us on Facebook Follow us on Twitter Follow us on Instagram