The estimation of population microdata by using data from small area statistics and samples of anonymised records

Census data can be represented both as lists and as tabulations of household/individual attributes. List representation of Census data offers greater flexibility, as the exploration of interrelationships between population characteristics is limited only by the quality and scope of the data collected. Unfortunately, the released lists of household/individual attributes (Samples of Anonymised Records, SARs) are spatially referenced only to areas (single or merged districts) with populations of 120 000 or more, whereas released tabulations are available for units as small its single enumeration districts (Small Area Statistics, SAS). Intuitively, it should be possible to derive list-based estimates of enumeration district populations by combining information contained in the SAR and the SAS. In this paper we explore the range of solutions that could be adapted to this problem which, ultimately, is presented as a complex combinatorial optimisation problem. Various techniques of combinatorial optimisation are tested, and preliminary results from the best performing algorithm are evaluated. Through this process, the lack of suitable test statistics for the comparison of observed and expected tabulations of population data is highlighted.