Skip to main content

School of Demography

  • Home
  • People
    • Academics
    • Visitors
    • Current PhD students
    • Graduated PhD students
  • Events
    • Seminar Series
    • Conferences
      • Past conferences
  • News
  • Students
  • Research
  • Contact us

Related Sites

  • ANU College of Arts & Social Sciences
  • Research School of Social Sciences
  • Australian National Internships Program

Administrator

Breadcrumb

HomeCreating 22 Million Synthetic Australians From Census Cross-tabulations
Creating 22 million synthetic Australians from census cross-tabulations

Unit records of persons, families and dwellings are needed for household microsimulations. These are available for a 1% sample of each census, but much larger samples with finer geographic subdivisions are needed for some purposes. For example, commercial decisions about the location of aged care centres require information about the numbers of persons needing care in local areas, and their ability to pay for the care.

This presentation will describe a program that is being written to create 22 million synthetic Australians, and their families and dwellings, from cross-tabulations from the 2011 census. Census under-reporting will be approximately corrected by using estimated resident populations at 30 June 2011. Dwellings will be located within the 2200 level 2 statistical areas. The program is being tested with data from the 1% sample from the 2006 census.

Shell databases for the 2001 census were released on April 30, with first release data for TableBuilder Pro due on August 7, and second release data on November 21. Details of the structure and deficiencies of census data may be of wider interest.

A sequential allocation method is being used to synthesize unit records. This involves allocating characteristics one at a time, using census cross-tabulations chosen for the purpose. Choosing the optimal order in which households are formed and characteristics are allocated is difficult. To avoid problems with data randomization, no cell of less than 4 is being used.

An attempt will be made to impute relationships between persons living in different households in Australia. Methods to do this have been used in Japan and Sweden, but it is not clear whether they will give credible results in Australia.

This work is intended to provide baseline data for a large household microsimulation model, simulating most of the attributes currently captured by the census, as well as other attributes such as wealth and diseases.


Richard Cumpston

Until September 2007 Richard was a director of Cumpston Sarjeant, consulting actuaries. In December 2011 he submitted a PhD thesis  titled New techniques for household microsimulation, and their application to Australia to the ANU school of actuarial studies. Since then he has been working on the project described above. His email is richard.cumpston@gmail.com, and his mobile 0433 170 276.

Date & time

  • Tue 08 May 2012, 3:30 pm - 5:00 pm

Location

Seminar Room A, Coombs Building, Fellows Road ANU

Speakers

  • Richard Cumpston