In Partnership with members of the NDA, the Australian Bureau of Statistics has produced Estimates of disability at the local government area level across Australia.
The Estimated areas are based on data collected from the 2003 Survey of Disability, Ageing and Carers, and the 2001 Census.
In early 2006, a series of workshops were conducted for National Disability Administrators (NDA) constituents to explain the quality and limitations of the small area estimates of disability released as an ABS consultancy. The workshop material and explanatory notes contained a series of diagnostic measures and broad assessments of the quality of the small area output.
Arguably the most important measure of the quality discussed at the workshops was the relative root mean square error (RRMSE), which represent the associated model error in predicting each small area estimate. The method for calculating the original RRMSEs excluded some components of total error that were subsequently shown to be sizeable. Recently an improved, though more computationally sophisticated method, was used to revise the RRMSEs. In the case of the disability estimates, the revised RRMSEs are up to 4 times larger than the original RRMSEs.
RRMSEs may be interpreted as follows: An RRMSE of less than 25% indicates that the associated small area estimate is of sufficient reliability for most statistical purposes. A small area estimate with an RRMSE of more than 25% but less than 50% is not considered to be sufficiently reliable for most purposes, and should be treated with caution while those with RRMSEs over 50% are too unreliable for most reasonable uses.
Many Release 1 estimates, previously with RRMSEs of around 5%, have been revised upwards to around 20%. While these still possess sufficient statistical reliability, it should be noted that they do not possess the high level of reliability indicated by the original synthetic RRMSEs.
Estimates contained in a number of tables, particularly those of Releases 2 and 3, have revised RRMSEs of over 25%. Had we initially produced random effects RRMSEs, these tables would not have been released, due to their insufficient statistical reliability. Only a small number of cells in specific tables have RRMSEs of over 50%.
Example - Revised RRMSE for Melbourne City
Table 2 of Release 2 showed that the LGA estimate of the proportion of children aged 4-9 years old with a disability in Melbourne city was 0.09 or 9%. The corresponding original RRMSE for this estimate was 7.4%.
This can be interpreted to mean that are 19 chances in 20 (or 95%) that the true, unknown value of this estimate of 9%, will fall between 7.7% and 10.4%.
The revised RRMSE for the proportion of children aged 4-9 years old with a disability in Melbourne City increases to 30.8%. This means that a more reliable indication of the quality of this estimate is that there are 19 chances in 20 that the true proportion actually lies in the range 4.8% and 16.1%. The wider confidence interval reflects the increased uncertainty about the true value of the proportion.
ACT - Under construction
New South Wales
Northern Territory - Not available
Queensland - Under construction
South Australia
Tasmania
Victoria
Western Australia
