A "PRACTICE MANUAL" FOR SMALL AREA ESTIMATION
The ABS, like other national statistical agencies, is facing increasing demands for estimates on smaller domains (smaller geographic areas, subpopulations, subindustries, finer time periods) than are supported by our surveys. Spread across the ABS is a good deal of experience with small area estimation based on a variety of techniques - ranging from simple prorating through to state-of-the-art Markov Chain Monte Carlo Bayesian methods, multilevel modelling and the like.
The Analysis Board recently commissioned a project that will distil the literature and the experience of statistical agencies into a "practice manual" that will provide guidance on such matters as:
Although the aim is to provide a roadmap of good practice (rather than to generate estimates addressing a particular statistical need), one or more demonstration projects may spin off from our work on the practice manual.
The project will be overseen by a board of economic and social statisticians and methodologists. Draft segments of the manual will be issued progressively during 2003.
For more information, please contact Ken Tallis on (02) 6252 7290.
- the classes of problems and datasets to which small area estimation can be applied;
- the spectrum of available methods;
- the circumstances in which each method might be most suitable for application in a statistical agency (including when a simple method might yield results almost as usable as a complex method);
- when explicit small area estimates might not be needed - and when the client's needs might be satisfied just as well by, say, a compendium of large area measures and small area indicators;
- how to define and assure the quality of small area estimates, and how to explain their quality and limitations to users; and
- where to find examples of each method and where to get help if you have been asked to develop small area estimates.