Directions for Household Survey Sampling Framework
The current broad sampling framework underpinning ABS household surveys has been in use for roughly 40 years. Household Survey Methodology (HSM) are currently investigating possible significant changes to the household survey sampling framework as an extension of the routine 5-yearly Monthly Population Survey (MPS) sample redesign hinging off the 2011 Census of Population and Housing. This timing allows for implementation of a new framework in the 2012/13 financial year. The changes being considered are driven by factors including:
- changes to the ABS geography classification structure, including the availability of a significantly smaller finest level of geography (mesh block) than current (Census collection district);
- relatively recent developments in sampling methodology and algorithms, i.e. balanced sampling via the cube method which shows potential non-negligible savings for MPS sample redesign;
- increasing demand for social statistics that require more sophisticated sample designs;
- a rebuild of sample management systems;
- increasing availability of new technology including mapping and geocoding software; and
- the need to find bookable efficiency gains for the organisation.
Currently the MPS, which consists of the Labour Force Survey (LFS) and its supplements, and Special Social Surveys (SSSs) are run off a single sampling framework. In this framework the Primary Sampling Unit (PSU), which is most often a Census Collection District (CD), is selected on the basis of MPS sample design but supports both MPS and SSSs. A block from each selected CD is used for MPS and another block, called the "parallel block", is used for SSSs. The sample design for MPS is area-based multi-stage, redesigned every five years focused on optimising design parameters towards LFS requirements. For both MPS and most
SSSs, designs are Equal Probability of Selection (EPS) within state, where clusters of dwellings from within blocks and CDs are systematically selected.
The current approach yields efficiencies in sample preparation activities (because the MPS and parallel blocks are set up at the same time), is easy to manage from an interviewer allocation perspective and controls effectively for overlap between collections. However, the approach has limited flexibility from a SSS sample design perspective, and given that the cost model underpinning the MPS sample design bears little resemblance to SSS enumeration costs, is probably sub-optimal for SSSs. In addition, recent SSS sample designs aimed at targeting specific subpopulations of interest have had to select sample independently of the parallel block framework. Recent investigations have also uncovered potential savings from balanced sampling for MPS sample design but these savings are not present, or at least relatively smaller, for SSSs.
As a result, two new sampling frameworks are being proposed as part of HSM's work, namely:
- An MPS master sampling framework (as it is currently), with an independent SSS sampling framework that will cover the majority of 'standard' SSS sample designs (i.e. for surveys of some specific subpopulations such as Indigenous people or pensioners, there will be no option but to have an independent sample design); and
- An MPS master sampling framework (as it is currently), and independent sample design and selection for all other surveys.
The first option allows greater flexibility in the design of SSSs, however it incurs additional sample preparation cost compared to the current method
(through decoupling of sampled areas and hence increase in sample preparation activities) and may not be flexible enough for some surveys. Overlap control is relatively easy to manage in this scenario, though will be more complicated than the current approach. The second option allows maximum flexibility for sample design activities for all surveys, but incurs the greatest additional sample preparation cost and presents the most challenging overlap control scenario. However this approach could yield greater overall efficiency if interviewer field costs greatly outweigh sample preparation costs, and the sample design could be more appropriately tailored to meet the needs of individual surveys.
Potential savings exist in reduced sample preparation activities such as "in-office" address coders for listing dwellings in selected areas and the new geography structure providing small "block type" units (ie mesh blocks) that are pre-defined, thus removing the need to divide a selected area into smaller groups of dwellings. Savings also exist in the potential use of balanced sampling for MPS selections as mentioned above (but not for detailed discussion in this article).
Broadly, HSM will be directing their immediate effort in the following areas:
- Establishing main advantages and disadvantages of the different household survey sampling frameworks proposed above;
- Reviewing other National Statistical Institute's (NSI's) approaches to household survey sampling frameworks;
- Examining potential cost modelling approaches and the data required for those cost models; and
- Establishing current and future demand for the ABS household survey program that will impact on sample design objectives.
For more information, please contact Justin Lokhorst on (08) 8237 7476 or firstname.lastname@example.org.