Small Area Estimation of LFS
Analytical Services Branch is undertaking a project to assess the feasibility of producing small area estimates of labour force at the Local Government Area (LGA ) level. The Labour Force Survey (LFS) produces monthly estimates by dissemination region. These have an average population of 250,000 persons, large enough to provide only a minimal level of sampling accuracy. The high sampling error on many dissemination region estimates, plus the fact that they tend to cross labour market boundaries, limit their use for government policy development and program evaluation.
This is not the first time the Australian Bureau of Statistics (ABS) has undertaken small area estimation for labour force statistics. In the mid 1980’s the ABS investigated the structure preserving estimation (SPREE) methodology developed initially by Purcell in 1979 to produce LGA level estimates of unemployment by age and sex. SPREE used iterative proportional fitting to adjust Centrelink (formally known as the Department of Social Security) counts of unemployment benefit recipients to LFS estimates. SPREE estimates of unemployment continue to be produced at the Department of Education and Work Relations (DEWR).
The current feasibility study follows the Office of National Statistics (ONS), UK approach of fitting a small area model to one month of LFS estimates. To date, the models we have fitted have been of fair quality but there are a number of obstacles we need to overcome before we are prepared to certify the quality of the output. These issues include:
- assessing the quality both in terms of the goodness of fit of the models and the reliability of the small area predictions themselves;
- whether we can easily improve on the quality of the model fit without resorting to more complex methods;
- the reliable prediction for off-census time points given that LFS covariates are not available for out of survey areas; and
- validation of the small area output using local expert knowledge.
A feasibility report will be produced in the coming months which will make recommendations on whether the output obtained so far is suitable for release and whether further work is likely to improve the estimates to a sufficient level.
For further information please contact Daniel Elazar on (02) 6252 6962.