|Page tools: Print Page Print All|
Improved estimation methodology for LFS family estimates
Since the 1970s, family estimates have been produced as a by-product of the Labour Force Survey (LFS). Family type data are derived from questions which establish the relationships between members of the selected households. The family estimates produced from LFS data classify families by type (e.g. couples without dependents, single parent families with children aged under 15), as well as by the labour force characteristics of the family (e.g. number of employed family members).
Recent improvements in the capture of information about the relationship between persons in sample households have enabled improvements to the estimation methodology for family data. Two distinct changes were made to the estimation methodology. First, more households selected in the LFS now contribute to family estimates, and second, a different method is used to calculate the sample weight of the contributing families.
Under the previous estimation methodology, the family weight was calculated as the harmonic mean of the LFS weights of each family member. A family weight was only produced for those families in which all members aged 15+ in their household contributed to the person-level LFS estimates. As a result of this, families in which one or more persons aged 15+ were out of scope (such as members of the Australian Defence Force) or did not respond to the LFS, did not contribute. Since the estimation method did not use independent population estimates (benchmarks) to compensate for these exclusions, family estimates were lower than they would have been if all families were included.
The new estimation method assigns a weight to all families in which the necessary relationship data are collected. The family weights are computed using a Generalised Regression (GREG) Estimator which uses demographic data about all individuals (including children) in the contributing sample families. Benchmarks for both the number of persons and households are used, so the weights are referred to as being 'integrated' at the person and household levels. Each household (and hence family) is assigned a weight such that the sum of the weights in each household benchmark class equals the household benchmark count, and when the household weight is assigned to each of its members, the weighted sum of individuals in each person benchmark class equals the person benchmark count.
The use of benchmarks enables the weighted sample to better represent the Australian population and ensures estimates are more comparable with family estimates from other ABS collections. Including more families in the estimates has also reduced bias. The sample error of estimates has been significantly reduced, with reductions in standard error of more than 40 per cent for those estimates which are a high proportion of the total number of families.
The first set of family estimates produced under the new estimation method used September LFS data and was released in October 2008. Also released at this time were an information paper on the new method (cat. no 6224.0.55.002) and estimates produced under the new method applied to historical data dating back to August 2004. Datacubes containing family estimates will now be released each month (cat. no. 6224.0.55.001) and can be downloaded from the ABS Website.
For further information, please contact Julian Whiting on (08) 8237 7362 or firstname.lastname@example.org.
These documents will be presented in a new window.