DISCUSSION
The use of the CDE Indigenous Mortality Study to assess the identification rate of Aboriginal and Torres Strait Islander deaths in death registration data has a number of benefits but at the same time it has limitations. First, the obvious and most substantial benefit is that it enables direct calculation of identification rates. That is, they are derived by directly comparing Indigenous status as reported according to death registrations and Census data for linked records, as opposed to indirect and modelled estimates.
Second, no assumptions were necessary to derive the identification rates from the CDE study, whereas a number of subjective judgements and assumptions were necessary to produce identification rates prior to the 2005-2007 period.
The limitations of the CDE Indigenous Mortality Study relate to three main factors. First, the derived Aboriginal and Torres Strait Islander deaths identification rates relate to a restricted time frame - from 10 August 2011 to 27 September 2012. It is not possible to accurately judge the appropriateness or otherwise of the derived Aboriginal and Torres Strait Islander deaths identification rates for past or future periods.
Second, there remains a relatively high level of unlinked Aboriginal and Torres Strait Islander death records. Of all the unlinked records, 22% were in Western Australia and 21% in the Northern Territory. While not unexpected given the relatively high Census undercount for Aboriginal and Torres Strait Islander Australians, there may be features or characteristics of the unlinked records that are quite different to the linked records and therefore may introduce some bias to the results. Sensitivity analysis (see Appendix 1) indicates this is likely to be small.
Third, as stated above, the propensities of Aboriginal and Torres Strait Islander identification as reported in the 2011 Census and 2011 PES are calculated for persons who matched to Census and PES. While PES is a sample representing the whole population, the current methodology implicitly assumes that the propensities based on the Census-PES match will apply for the death registrations linked to the Census. Appendix 1 provides some indication of the sensitivity of the estimates of Aboriginal and Torres Strait Islander life expectancy at birth to assumptions made when calculating it.
A further assumption relates to the application of identification rates, which assumed uniformity by age and sex - except for the headline Australia estimate (which used three broad age groups).
In spite of these limitations, the CDE Indigenous Mortality Study clearly shows the need to adjust for underidentification in Aboriginal and Torres Strait Islander death registrations.