Bayesian multivariate microaggregation under the Hellinger’s distance criterion. Research.pdfVIP

Bayesian Multivariate Micro-Aggregation Under the Hellinger’s Distance Criterion G. KOKOLAKIS and PH. NANOPOULOS National Technical University of Athens, Greece and CEC, Eurostat, Luxembourg Abstract: The technique of micro-aggregation is widely used in Official Statis- tics to reduce the risk of identity disclosure. Micro-aggregation implies loss of information. In order to minimize the loss of information while protecting con- fidentiality of individual data, we present here a Bayesian method of optimal multivariate micro-aggregation by minimizing the Hellinger’s distance criterion between the two posteriors based on the original and the micro-aggregated mul- tivariate normal data. Keywords: STATISTICAL CONFIDENTIALITY, DATA MASKING, HELLINGER’S DISTANCE, IDENTITY PROTECTION, MULTIVARIATE MICRO-AGGREGATION, PARTITIONING. 1. INTRODUCTION One of the main problems that National Statistical Offices face when disseminating infor- mation to the public is that they are not allowed to deliver, even anonymous individual records, if there is any possibility of direct or indirect identification of individual units. This is particularly true for business data. Their task is to protect the individual data from what Duncan and Lambert (1989) have called 
    ! #$  . This basic requirement of Official Statistics frequently has negative implications in pursuing statisti- cal studies and thus researchers complain about the lack of microdata. Duncan and Pearson (1991) state that “a number of social science research and public policy studies could be pursued if the present tension between access and confidentiality were better resolved. For more on this topic see also Fienberg (1994). In order to resolve the above