This issue contains two articles:
- Reducing provider burden in the Quarterly Business Indicators Survey
- Estimating the Aboriginal and Torres Strait Islander resident population
Features important work and developments in ABS methodologies
This issue contains two articles:
The ABS has been exploring options to reduce provider burden in one of our largest business surveys, the Quarterly Business Indicators Survey. This is part of our broader priorities to reduce provider burden and make greater use of administrative data sources.
Currently, the survey uses a simple estimation and weighting approach which does not significantly utilise administrative data.
Two options to reduce provider burden were considered:
Composite regression estimators are used in surveys that are focused on movement estimates which have appropriate auxiliary benchmark data. This type of estimator is already used in the ABS Labour Force Survey and other national statistical offices. The estimator is an extension of the generalised regression estimator which has increased benefits in reducing the sampling error on movement estimates.
Broadly, the method combines data collected in both the previous and current quarter to produce the current quarter’s estimates. There are efficiencies gained by exploiting the high correlation between overlapping samples and calibrating to administrative data population benchmarks.
The direct substitution of administrative data had various caveats. To maintain quality and coherence, administrative data sources have to align with existing survey concepts, and be available at similar timeframes and with enough coverage that we can accurately impute or adjust for missing data.
Research to date has shown some administrative data sources are better suited for use in this survey than others. Some sources had quirks, lower coverage, and greater timing risks which would have resulted in compromises to quality to achieve the desired burden reduction. Other sources looked more promising in terms of timing and coverage of the target population, though would only apply to a subset of the desired burden reduction. Further research is required before the administrative data sources can be used for direct substitution in the survey.
Based on the investigation undertaken to reduce provider burden in the survey, the aim is to implement composite regression estimator in mid-2023. We anticipate this to result in a sample reduction of up to 25% (4,000 businesses) whilst maintaining a similar level of output quality to that currently. We are in the process of finalising the methodology (by testing different calibration classes and parameters), as well as considering system implications and coherence with existing published estimates.
For more information, please contact Joey Srinkapaibulaya.
Each Population Census gives the ABS an opportunity to update estimates of the Aboriginal and Torres Strait Islander resident population, both at a national level and for regions of Australia. The process combines Census counts with estimates of the number of persons that have been missed or double counted by the Census (obtained through results from the Census Post Enumeration Survey, or PES). This is known as an undercount adjustment.
Some state and territory estimates of net undercount for the Aboriginal and Torres Strait Islander population have relatively high standard errors, due to the limited size of the PES sample. To account for this, the Aboriginal and Torres Strait Islander resident population estimates are obtained by smoothing PES-based net undercount estimates towards a stable predicted value. Since 2011, the ABS has applied an Empirical Bayesian model to undertake this smoothing.
The smoothing is applied to a set of custom regions, rather than at state and territory levels. Creating custom regions produces more homogenous regions with respect to Census undercount, leading to higher quality smoothed estimates. In 2021, the Empirical Bayesian model smoothed the Aboriginal and Torres Strait Islander undercount adjustments for eighteen regions. These custom regions reflected the relevant population sizes, and usually separated the capital city from one or two other regions within each state or territory.
A prediction of the proportion of Aboriginal and Torres Strait Islander persons missed in the Census is modelled for this set of regions based on a logistic regression model. The model incorporates Census variables found to be significant predictors of Census undercount, and includes
The Bayesian modelling approach is then applied to smooth the net undercount estimates obtained from PES towards these predictions plus a common intercept value. The amount of smoothing used is calculated using maximum likelihood estimation, with the level of smoothing in each region dependent on the size of the PES standard error. Regions with higher standard errors require more smoothing.
Following the application of the smoothing, the ABS conducts a set of quality checks. This ensures the Empirical Bayesian model has produced high quality outputs to be used in estimating the Aboriginal and Torres Strait Islander resident population.
The results of the smoothing and the preliminary population estimates were published on 21 September 2022 in Estimates of Aboriginal and Torres Strait Islander Australians.
For more information, please contact Soraya McPhail.
Please email firstname.lastname@example.org to:
Alternatively, you can post to:
Methodological News Editor
Australian Bureau of Statistics
Locked Bag No. 10
Belconnen ACT 2617