IMPROVED ESTIMATION METHOD


IMPROVEMENTS

From October 2008, the method of producing family estimates from the LFS will be improved in the following ways:

• The conceptual scope of the family estimates will widen to include households containing permanent members of the Australian defence forces.
• A wider range of families in the LFS sample will contribute to the family estimates.
• An improved weighting method will be employed, which makes use of independent population benchmarks (of persons and households) to ensure the estimates more closely reflect the Australian population.

SCOPE OF THE FAMILY ESTIMATES

Australian defence force personnel to be included

The current method of producing family estimates begins with those records used in person-level processing. The aim of person-level processing is to produce employment estimates of the civilian population of Australia. Permanent members of the Australian defence forces are explicitly excluded from the scope of LFS person-level estimates. Because LFS family estimates were derived from the person-level estimates, the families of Australian defence force personnel have until now been excluded from the scope of LFS family estimates.

Under the improved method, the scope of the improved family estimates is independent of person-level processing. The improved method will provide estimates of all Australian families who usually live in private dwellings, including the families of Australian defence force personnel.


Non-private dwellings and other continuing exclusions

Persons in non-private dwellings (for example hotels, hospitals and camping grounds) are in scope of LFS person-level estimates, but will continue to be excluded from the scope of LFS family estimates. The LFS does not collect relationship information in non-private dwellings. Most people selected in non-private dwellings in the LFS are staying there temporarily, and usually reside in a private dwelling.

A number of other groups excluded from ERP will continue to be excluded from the scope of LFS family estimates. A household will be excluded from family estimates if any of the usual residents aged 15 years and over are:

• Non-Australian diplomats, non-Australian diplomatic staff or non-Australian members of their household;
• Short term overseas visitors, whose usual residence is outside of Australia and who are staying in Australia for less than 12 months; or
• Members of non-Australian defence forces stationed in Australia or their dependants.

WIDER RANGE OF FAMILIES INCLUDED

Under the improved method, data from all private dwellings will contribute to family estimates, as long as all demographic and family type variables are available. This will provide a more complete picture of family characteristics from the LFS than has been available in the past.

Since the introduction of Computer Assisted Interviewing (CAI) in the LFS in August 2004, basic demographic information has been captured for all persons in selected dwellings including those out on scope of the LFS. Since July 2006, additional information has been collected (for example, whether persons aged 15-24 were in full-time education) for all usual residents of private dwellings, including those out on scope. These changes allow all households in private dwellings to be coded to family type, and used in the production of family estimates.

With the introduction of the improved method in October 2008, information from the following groups will contribute for the first time to LFS family estimates:

• Households containing permanent members of the Australian defence forces;
• Households containing usual residents of private dwellings who fail to meet normal LFS selection rules (for example, because they are away from their usual residence for more than six weeks); and
• Households which did not fully respond to the survey (as long as all demographic and family type variables are available).

Discrete Indigenous communities

Data on relationships within households in discrete Indigenous communities are not available from the LFS, due to the complexity and cost of collecting this information. Therefore, relationship data from discrete Indigenous communities is not available for use in the production of LFS family estimates.

However, the scope of LFS family estimates is all private dwellings, including private dwellings in discrete Indigenous communities. The use of independent population benchmarks in weighting (explained below) compensates somewhat for the lack of household relationship information in discrete Indigenous communities, by ensuring that the family estimates are consistent with benchmark counts of persons and households for a given area (e.g. Northern Territory).

Any benchmarking process assumes that the unobserved data are represented by the observed data. The improved LFS family methodology assumes that families in discrete Indigenous communities have similar characteristics to all other families in the same geographic area (see Improved Weighting Method below for the geographic areas used in benchmarking). This assumption should be kept in mind, particularly when interpreting data for the Northern Territory, where approximately 20% of the population lives in discrete Indigenous communities. If families in discrete Indigenous communities have different characteristics to families in the rest of the Northern Territory, for example if families are larger, then the number of families in the Northern Territory will be overstated. The improved LFS estimate of families in the Northern Territory are likely to be much closer to the true number of families than the current LFS estimate (see the comparison with other ABS sources of families data in the next section of this publication), but some characteristics of those families may be less reliable.


IMPROVED WEIGHTING METHOD

The improved weighting method to be introduced from October 2008 makes use of independent population estimates, known as benchmarks. The method makes use of benchmarks of both persons and households. The use of benchmarks has three benefits:

• It ensures the estimates are more representative of the Australian population;
• The estimates are more comparable with estimates from other ABS collections; and
• Sampling error is considerably reduced.

Person benchmarks

The estimates of people used in the family estimates will be weighted to match to independent counts of persons for each cross-classification of the following class variables:

• geographic area - State capital city (e.g. Sydney), balance of State (e.g. New South Wales, excluding Sydney), Northern Territory and Australian Capital Territory (14 classes);
• sex (2 classes); and
• age group - 5 year age groups between 0 and 74, 75 years and over (16 classes).

The combination of these three variables gives 448 benchmark classes (14 x 2 x 16 = 448). The weighted total of persons within each benchmark class is constrained to match the population count in that class. For example, the weights of 15-19 year old males who live in Sydney will be adjusted so that they total the estimated resident population of 15-19 year old males in Sydney.


Household benchmarks

The number of households determined from the processing of family estimates will be weighted to match the independent counts of households in Australia for each cross-classification of the following class variables:

• geographic area - State capital city (e.g. Sydney), balance of State (e.g. New South Wales, excluding Sydney), Northern Territory and Australian Capital Territory (14 classes); and
• household composition - the number (1, 2 or 3+) of usual residents who are aged 15 years or over, and the number (0, 1 or 2+) of usual residents who are aged under 15 years (9 classes).

The combination of these two variables gives 126 benchmark classes (14 x 9 = 126). The weighted total of households within a benchmark class is constrained to match the total number of households in that class.

For example, the weights of households in Sydney with one usual resident aged 15 years or over and one usual resident aged under 15 years will be adjusted so that they total the estimated number of households in Sydney with the same household composition.

The current household benchmarks used in the derivation of family estimates are based on the 2001 Census of Population and Housing. The incorporation of information from the 2006 Census of Population and Housing and a review of the methodology for producing household benchmarks is likely to result in revisions to the household benchmarks and in turn the family estimates.


Computing the family weight

Each family within the scope of family estimates will be assigned an expansion factor or weight, using a form of 'Generalised Regression' (GREG). The outputs of this process are household weights, which are set to the family weights and person weights for the families and individuals in the household. The sum of the final weights within each benchmark class are constrained to equal the benchmark count at both the person level and household level.

Estimates for each characteristic of interest are then obtained by summing the weights of families in the sample with that characteristic.


SUMMARY

To reiterate, the improved estimation method will, through the inclusion of a wider range of families and the use of independent population benchmarks, result in family estimates that more closely reflect the Australian population.

The main limitation of the improved estimation method is the necessary assumption that families in discrete Indigenous communities have family characteristics similar to other families in the same area. This may result in some overstatement of families in the estimates.