|Page tools: Print Page Print All RSS Search this Product|
CHAPTER 18. METHODS USED IN ABS HOUSEHOLD SURVEYS
Special dwelling framework
18.16 The special dwelling framework is also stratified geographically, though at a broader level than the private dwelling framework. In many cases the demographic, social and labour force characteristics of the occupants of special dwellings are not typical of the population residing in private dwellings, and therefore it is necessary to sample special dwellings separately by placing them in separate strata within each geographic (sample design) region. This provides for more effective samples of persons within special dwellings and private dwellings, and the flexibility to select some samples which exclude all or some special dwellings, or to select samples in which special attention is paid to persons residing in particular special dwellings.
Indigenous Communities Framework
18.17 Communities on the ICF are stratified geographically by state. Finer level stratification for the ICF was not feasible as the number of CDs in each state was not sufficient to enable the creation of multiple strata within any state/territory.
POPULATION SURVEY MASTER SAMPLE
18.18 As discussed above, the Population Survey Framework is used to select a sample which forms the Population Survey Master Sample. The Population Survey Master Sample comprises a sample of PSUs taken from the private dwelling framework, special dwellings, and Indigenous communities from the Indigenous Communities Framework. Most household surveys conducted by the ABS use samples drawn from the Population Survey Master Sample.
18.19 To satisfy the varying sampling and collection requirements of the many surveys using the Population Survey Master Sample, the sample has been designed to be as flexible as possible. Although the LFS, SSS and other ABS household surveys all have separate samples of dwellings, they are all selected from the Population Survey Master Sample using the same procedures. In many cases, much of the field work involved in setting up the sample is common to several surveys.
18.20 The Population Survey Master Sample is reselected and redesigned every five years following the Census of Population and Housing (see paragraph 18.41).
18.21 A multistage (footnote 4) area sample design is used to draw the sample of private dwellings from the list of PSUs on the private dwelling framework. For most areas there are three stages of sample selection: the first stage involves the selection of a sample of PSUs (each PSU generally equates to a CD); the second involves the selection of blocks within the selected PSUs; and the third involves the selection of dwellings within the selected blocks. However, for areas in the ex-metropolitan non-SRA strata (i.e. least remote, remote and very remote areas) there are four stages to sample selection: the first stage involves the selection of a sample of PSUs (each PSU is a group of contiguous CDs); the second involves the selection of a sample of CDs from the selected PSUs; the third involves the selection of blocks within the selected CDs; and the fourth involves the selection of dwellings within the selected blocks.
18.22 The use of cluster sampling ensures that the 'final sampling units', i.e. the dwellings selected in the sample, form groups within which the dwellings are close together. This reduces the cost of compiling dwelling lists, and the amount of interviewer travel between selected dwellings. Highly clustered samples, where a large number of dwellings are selected from each of a small number of small areas, produce the lowest operational costs, but the results are less reliable and less representative of the population at large (i.e. have higher standard errors). A key feature of the design is achieving an acceptable balance between cost and accuracy.
18.23 A multistage design is also used to draw the sample of special dwellings from the special dwelling framework. A sample of special dwellings is selected at the first stage and dwelling units (rooms, beds, caravan sites etc.) at the second stage.
18.24 As with private and special dwellings, a multistage design is used to draw the sample of Indigenous communities from the ICF. A sample of Main Communities is selected at the first stage, and at the second stage a sample of outstations associated with the selected Main Communities are taken. Each selected outstation is them completely enumerated. A list is maintained of non-community dwellings within the CDs associated with each selected Main Community. These non-community dwellings are turned into clusters according to the number of clusters in the Main Community. A cluster sample is then taken of these non-community dwellings. Special dwellings in the selected CDs are regarded as non-community dwellings within the ICF.
18.25 The allocation of sample between the states is a compromise between accurate national estimates and useable estimates for the smaller states. As a result, the smaller states (and the territories) have relatively high sampling fractions. The sampling fraction is uniform within states, reflecting the fact that, at the time the methodology was originally devised the state was the smallest geographic area for which estimates were usually published. Some allowance for the higher costs of sampling in non-metropolitan areas is made in the increased clustering of the sample in those areas.
18.26 Systematic sampling is used as the selection method for private dwelling, special dwelling and Indigenous community samples. Geographic location is used as the ordering variable for this systematic selection.
18.27 As discussed in paragraph 18.21, for most areas there are three stages to sample selection. At the first stage, PSUs are selected systematically with probability proportional to size (in terms of number of dwellings) and drawn independently from within each state by area type. At the second stage, blocks of 25-50 dwellings are formed and two blocks are selected, again with probability proportional to size. One block is used for the LFS sample and the other for a parallel sample. The parallel sample is used for the SSS, and for other household surveys where probability based samples are required. At the third stage, all dwellings in the selected blocks are listed and a 'cluster' of dwellings - comprising the final selection units - is selected using systematic random sampling. This process limits the need for a listing of dwellings to only those in the selected blocks.
18.28 For private dwellings, the selection of sample units at each stage (other than the last) results in a probability of selection proportional to the approximate number of dwelling units the CDs or blocks contain. At the last stage, in which dwellings are selected, each dwelling within a selected block has the same chance of selection.
18.29 The sampling fractions and selection procedures result in a self-weighting sample which guarantees every dwelling in the same state or territory the same chance of selection.
18.30 A systematic sample of Special Dwellings (SD) is selected from the special dwelling list, with probability proportional to size. The SDs are sorted by region, PSO type, area type and descending PSO occupancy. A systematic probability proportional to size sample of SD's is selected from the SD list independently within each state. Within states, SDs with a measure of size greater than 4 times the state skip interval are selected with certainty and placed at the top of the list. In least remote, remote and very remote areas, the recruitment of interviewers is difficult and the travel costs are high. Therefore, SDs in least remote, remote and very remote strata are selected only if they lie in a PSU selected in the PD sample or in a least remote urban PSU adjacent to a selected least remote rural PSU. This is known as "urban rural association". Within selected SDs, only usual residents (UR) are enumerated.
18.31 Average occupancy is used as the 'measure of size' of the special dwelling for selection purposes.
18.32 The selection methodology for Indigenous Communities involves forming 'sets', comprising each Main Community and its associated outstations. At the first stage, Main Communities are selected systematically with probability proportional to size of the 'set' and is referred to as the 'measure of size'. This selection methodology gives bigger 'sets', for which a reasonable workload for the set is assured, more chance of being selected.
18.33 At the second stage, a systematic sample is taken of the outstations associated with each selected Main Community. All dwellings within each selected outstation are completely enumerated. All outstations within a set have an equal probability of selection.
18.34 The measure of size used in the first stage is dependent upon the number of dwellings in the Main Community, as well as the number of 'live' outstations associated with it. A 'live' outstation is defined as an outstation that was populated at the time of the 2001 Census of Population and Housing, or Community Housing and Infrastructure Needs Survey (CHINS) 2001 (if no Special Indigenous Form (SIF) was required). The selection strategy puts an upper limit of two 'live' outstations selected per Main Community.
18.35 Once the measure of size is determined, it becomes the basis for determining the number of clusters formed from both the Main Community and the non-community dwellings. Therefore, the number of non-community dwelling clusters within each 'set' is equal to the measure of size for that 'set'. Clusters are selected using the same random start as that for the Main Community and rotated the same as the clusters for the Main Community, therefore systematic sampling is also used for the selection of the non-community dwelling clusters within the CDs associated with the Main Community in the selected 'set'.
18.36 This selection methodology applied to the ICF ensures that every person on the frame has the same chance of being selected. It is called Equal Probability of Selection (EPS). This also holds for the selections from the PD frame.
Adjusting for growth
18.37 The Population Survey Master Sample is reselected once every five years (see paragraph 18.40), with the sample being used up progressively by ABS household surveys over each intervening five year period. Regular checking of the dwellings listed for each selected block, combined with the use of systematic random sampling at the last stage, ensure that all dwellings have the correct probability of selection for all surveys during the five year period (that is, the sample automatically adjusts for growth).
18.38 For the private dwelling framework, special 'growth revision' procedures are applied in areas where pockets of substantial growth have occurred (e.g. new housing subdivisions). Strata where substantial non-uniform growth has occurred are identified (using building approvals data) and each CD within them is represented twice on the frame, once for the dwellings in the CD as at the last Census and once for the growth dwellings. This means that the growth CDs can be selected twice in the sample. When revision of the CD sample is necessary due to growth in the stratum, additional clusters are allocated to the stratum with extra selections made from these clusters. The number of clusters to be allocated will depend on the amount of growth in the growth CD's of that stratum. Additional clusters should be allocated to a growth stratum according to the value of the total growth in the growth CD's, where total growth in each CD is supplied by Central Office of the ABS.
18.39 Lists of special dwellings are updated from Census information, previous lists of special dwellings, local knowledge, commercial and other directories, and field work (ie non-sample sources or independent source feedback). Special dwellings not on this list are accounted for in the private dwelling sample.
Predetermined Growth Strata
18.40 In some areas of Canberra and Darwin it is known that significant growth will occur during the life of the sample. Special growth revision procedures are used for these areas to simplify and minimise the amount of effort required to account for the growth. The CDs which constitute a growth stratum are typically in areas of rapid growth, contain large plots of vacant land on which the new dwellings are to be built and are usually geographically contiguous with similar CDs. That is, the growth stratum is not used for a CD which will grow significantly but which is bounded by fully built-up CD's. The selection of dwellings from growth strata does not include a CD stage of selection. The procedure differs depending on whether maps of the planned building lots are available.
18.41 The ABS reselects the Population Survey Master Sample every five years, using preliminary data from the Census of Population and Housing. For the ICF, data from CHINS is also used to determine other community characteristics, as well as supplementing Indigenous community data from the Census to reduce the possibility of undercoverage in matching between the PD framework and ICF. These Censuses provide valuable information to update the frame for selections. This reselection ensures that the household survey samples continue to accurately reflect the distribution of the Australian population. As well, the opportunity exists to examine the overall design of individual household surveys to ensure that they remain efficient and cost-effective.
18.42 Household survey estimates are generally calculated using calibration estimation techniques.
18.43 Estimates of the population produced from household surveys are calculated in such a way as to add up to independently estimated counts (benchmarks) of the population. For the LFS these benchmarks are based on Census of Population and Housing data, adjusted for under-enumeration and updated for births, deaths, interstate migration, and net permanent and long term migration. Benchmarks have been developed for state/territory of usual residence, part of state of usual residence (for example, capital city, rest of state), age and sex. Each cross-classification of these benchmark variables is known as a benchmark cell. Revisions are made to benchmarks after each Census of Population and Housing, and when the bases for estimating the population are reviewed.
18.44 Other household surveys use various combinations of benchmark variables to produce benchmark cells. Some surveys use supplementary information (such as LFS estimates), referred to in this context as pseudo-benchmarks, to supplement independent demographic benchmarks based on Census of Population and Housing data. Household surveys may use calibration methods to incorporate other auxiliary information on target populations into estimates - for instance benchmarks for the Indigenous population or the population of private households.
18.45 For most household surveys, a non-response adjustment is performed implicitly by the estimation system, which effectively imputes for each non-responding person on the basis of all responding persons in the same post-stratum. This adjustment accounts for both full non-response and non-response for individual questions.
18.46 The response rate usually quoted for ABS household surveys is defined as the number of fully responding households divided by the total number of selected households excluding sample loss. Examples of sample loss for household surveys include: households where all persons are out of scope and/or coverage; vacant dwellings; dwellings under construction; dwellings converted to non-dwellings; derelict dwellings; and demolished dwellings.
18.47 For further details contact the Labour Market Statistics Section, on Canberra (02) 6252 7206, or email firstname.lastname@example.org.
1. Institutions are defined as: hospitals and homes (including general homes, other hospitals, convalescent homes, homes for the aged, retirement homes, homes for the handicapped and orphanages), and prisons. Institutionalised persons are defined as all persons selected in institutions, apart from live-in staff who do not usually live in a private dwelling. <back
2. Boarding school pupils are defined as all pupils selected in boarding schools. <back
3. Note there may be more than one household associated with a dwelling. <back
4. Multistage sampling is an extension of cluster sampling - see Chapter 17 for further information. <back