• Summary
• Downloads
• Explanatory Notes
• Related Information
• Past Releases

Industrialising Small Area Estimation at the ABS

ABS surveys are designed to achieve accurate estimates using design-based methods at broad levels, for example, state and national levels in household surveys, or industry divisions in business surveys. At finer levels of geography, or detailed industry classifications, there is usually insufficient sample for reliable estimation. However there is a strong government demand for these small area statistics for planning, policy development and resource allocation. Small area estimation methods are one way in which this demand might be addressed.

In small sample situations, model-based methods are often able to provide estimates of acceptable accuracy. Statistical models are used to borrow information from other areas or time points via auxiliary data related to the target variable. This auxiliary data may include Census or external administrative data sources. In the ABS, model-based, small area estimates have been produced for a few external consultancies and been published as a ‘complementary’ product in a couple of cases. Typically the methods and processes involved are time-consuming and bespoke, but lead to high quality estimates.

It is envisaged that in the future small area estimates may be released as official statistics. A number of challenges arise, especially the need for systems and processes that are ‘mass customisable’. This means the ability to output a large number of estimates quickly, as well as the flexibility to implement different methods for different situations. To this end we have initiated research to examine a number of alternative model-based methods. A focus has been on weight-based approaches as these are particularly suited to production-efficiency. These include the model-based direct estimator of Chandra and Chambers¹ and the re-weighting method described by Schirm and Zaslavsky². Work is also progressing to understand and map the business process and business requirements for a future small area system.

Other challenges to address include achieving a right balance between efficient ‘black box’ processes and the need for quality assurance and expert technical input. Good knowledge management built into intelligent systems will be important here. The ability to produce reliable, production-efficient mean square error estimators for small area estimates is also a key area of research.

The task of industrialising small area estimation is a challenge, in particular attempting to simplify and streamline a process that is by nature very complex is difficult. Work to date indicates good prospects of success.

¹ Chandra, H and Chambers, R. (2009). Multipurpose weighting for small area estimation. Journal of Official Statistics, 25, 1-18.

² Schirm, Allen L. and Alan M. Zaslavsky. Reweighting Households to Develop Microsimulation Estimates for States. 1997 Proceedings of the Section on Survey Research Methods. Alexandria, VA: American Statistical Association, 1997.

Further Information
For more information, please contact Sean Buttsworth (02 6252 5174, sean.buttsworth@abs.gov.au) or Peter Radisich (02 6252 6731, peter.radisich@abs.gov.au).

The ABS Privacy Policy outlines how the ABS will handle any personal information that you provide to us.