An evaluation of the combinatorial optimisation approach to the creation of synthetic microdata

Abstract

Population microdata comprise a list of households and individuals, each with an associated list of characteristics. Unfortunately, for Britain no small-area microdata exist that offer the broad range of demographic and socio-economic variables contained in the national census. There are a number of approaches to the reconstruction of such spatially detailed microdata, including data fusion, synthetic reconstruction (chain imputation) and reweighting. One variant of the reweighting approach involves the selection of a combination of households from the 1% household SAR that best fits known small-area constraints (the published census tabulations). In this paper the implementation of this ‘combinatorial optimisation’ technique is more thoroughly examined. First the combinatorial optimisation process is reviewed. Then a number of methodological innovations, designed to improve the accuracy and consistency of resulting outputs, are reported. Subsequently the problems of evaluating the outputs are discussed and a new strategy is outlined for assessing the quality of synthetic microdata. The strategy proposed is generally applicable to all such data, irrespective of their means of generation. The paper goes on to provide an extensive assessment of the quality of synthetic microdata produced using the combinatorial optimisation approach. This represents the first time that such an evaluation has ever been undertaken. The results highlight the degree to which such data may be able to meet specific user needs. The paper concludes by offering an illustration of the ‘added value’ that can be obtained by combining information from public-use microdata with published small-area tabulations. Copyright © 2000 John Wiley & Sons, Ltd.