Extending the MaCSim approach using similarity weight matrix to assess the accuracy of record linkage
share
Record linkage is the process of bringing together the same entity from overlapping data sources, both administrative and substantive while removing duplicates. Statistical, health, government and business organisations often link administrative, survey, population census and other files to create a new file with complete information for comprehensive analysis. It is also crucial to assess the accuracy of the linked file in order to make valid inferences based on the combined data. Interestingly, there has been limited work on this issue of record linkage accuracy to date. In our first paper, we proposed a Markov Chain based Monte Carlo simulation approach (MaCSim) for assessing linkage accuracy and illustrated the utility of the approach using a synthetic dataset provided by the Australian Bureau of Statistics (ABS) based on realistic data settings. MaCSim utilizes two linked files with known true match status to create an agreement matrix and then simulates the matrix using a proposed algorithm developed to generate re-sampled versions of the agreement matrix. In this paper, we aim to improve this method by calculating a similarity weight to create the agreement matrix. This weight allows partial agreement of the linking variable values for record pairs in the form of a similarity weight. To assess the average accuracy of linking, correctly linked proportions are investigated for each record. Test results on this extension of the MaCSim approach using the similarity weight concept show higher accuracy of the assessment of linkages.
READ FULL TEXT