Generalised regression estimation for ABS Business Survey
by
James Chipperfield
Australian Bureau of Statistics

The availability of Business Activity Statement (BAS) data collected by the Australian Taxation Office (ATO) provides the Australian Bureau of Statistics (ABS) with an opportunity to improve the efficiency of sample designs and estimators for its business surveys. The ABS hopes to exploit this opportunity by developing methods to improve the efficiency of estimation which make use of multiple auxiliary variables. Furthermore, the potential of this opportunity will improve with the alignment of BAS and ABS survey-collected information. This presentation introduces the generalised regression estimator and briefly mentions some details of the proposed methods of implementation within the ABS. The main focus of this presentation is the development of a variance estimation methodology for generalised regression estimators for ABS business surveys. The first stage of this development was a literature review of alternative variance estimation methods, and the evaluation of these alternatives against the constraints of ABS business survey designs and a generalised estimation system within a national statistical organisation. The second stage involved developing theory to meet the ABS's core requirements of a generalised estimation system, including point-in-time and movement estimators. The third stage involved carrying out an empirical evaluation of alternative Bootstrap variance estimators and of a variant to the Balanced Repeated Replication estimator. Lastly, the presentation recommends a Bootstrap variance estimation method and discusses some of the outstanding issues and the actions that are being undertaken to address them.

Date received: April 10, 2002