NATSISS 2008 sample design uses meshblocks
The Indigenous population of Australia is roughly 2.3 per cent the total population. In survey terms, the Indigenous population is a rare population. There is no list of Indigenous persons within Australia, so conducting a survey of the Indigenous population can be a challenging and costly exercise.
ABS Indigenous surveys have two components. The non-Indigenous Community component uses a screening mechanism to assist in selecting the sample. Census Collection Districts (CDs; a geographic unit) are selected and dwellings within the CD are approached (or 'screened') to determine if an Indigenous person resides in the dwelling. For example, under this design, screening approximately 230,000 dwellings would result in an expected sample size of 5,100 responding Indigenous dwellings, equating to roughly one responding Indigenous dwelling per forty-five dwellings screened.
Recently, a new geographic unit called a meshblock has become available. Meshblocks are generally much smaller than CDs, being roughly one-seventh the size of a CD. In designing the sample for the 2008 National Aboriginal and Torres Strait Islander Social Survey (NATSISS 08), the Household Survey Methodology Section investigated the possibility of using meshblocks to improve the efficiency of the NATSISS 08 sample design. It was hoped that using meshblocks would result in a large reduction of screened dwellings for the same level of survey accuracy.
The proposed sample design was to use meshblocks as the secondary stage of selection. That is, CDs are selected; then within CDs, meshblocks are selected and screened for Indigenous dwellings. Meshblocks expected to contain no Indigenous dwellings (from Census data) were removed from the coverage of the survey. Under this design, 5,100 responding Indigenous dwellings could be expected to be obtained by screening approximately 137,000 dwellings - equating to roughly one responding Indigenous dwelling per twenty-seven dwellings screened. However, due to the high mobility rate of Indigenous persons, it is expected that some of the meshblocks removed from the coverage of NATSISS 08 will contain Indigenous persons by the time NATSISS 08 is in the field. The level of undercoverage (i.e. proportion of Indigenous persons excluded from the coverage of NATSISS 08) was estimated to be approximately 8.7 per cent across Australia, but was particularly high in certain parts of Australia, such as Victoria, where the undercoverage was estimated to be 17.1 per cent. This level of undercoverage was not acceptable.
Meshblocks expected to contain no Indigenous dwellings (from Census data) can be included in NATSISS 08 to reduce the undercoverage, however their inclusion increases the level of screening. Subsampling these meshblocks results in a lower increase in the level of screening and reduces the undercoverage, but increases the sample error, decreasing the survey accuracy. An acceptable balance of screening level, undercoverage and sample error is required and was achieved by careful adjustment of these factors.
The resulting NATSISS 08 sample design is expected to have undercoverage of 4.4 per cent and to obtain 5,100 responding Indigenous dwellings from a screened sample of approximately 154,000 dwellings, being roughly one responding Indigenous dwelling per thirty dwellings screened. This equates to a saving of 76,000 dwellings screened for the same expected sample size.
For further information about the sample design of NATSISS 08, please contact Jennifer Webb on (02) 6252 5944.