|Page tools: Print Page Print All
The investigations undertaken by the ABS have identified two main concerns with the use of indirect methods. First, for both the Bhat and Hill methods the resulting estimates of coverage of Indigenous deaths for 2001-2006 are markedly lower (20-25 percentage points) than the coverage rates for the 1996-2001 period, implying that coverage of Indigenous deaths in death registration data has worsened considerably over the last five years. Such a result would seem implausible when compared to the observed data. Second, analyses undertaken by ABS and others (Barnes et al., 2008) have highlighted that the life expectancy estimates are very sensitive to the quality of the population estimates at the respective end points. Errors in the age distribution and level of either population estimate can result in very different life expectancy outcomes.
Based on the analysis of the 2006 data, the ABS is of the view that the indirect demographic methods are not suitable to the Australian situation and no longer recommends their use for the purpose of compiling Indigenous life expectancy estimates.
The direct demographic method is the method used for deriving life expectancy for the Australian population without any adjustment to deaths data. It takes the population at a point in time and the registered deaths around that period (e.g. the average of three years to reduce the effect of year-to-year variation) to compile life tables and life expectancy estimates. The method assumes good quality population estimates and assumes complete coverage of deaths.
The direct method can also be applied with a prior adjustment to the deaths or population data based on assumptions regarding coverage. Previously the ABS only had the indirect methods for deriving coverage estimates in respect of deaths. However, in 2006 a more direct measure of coverage was possible using results from the Census Data Enhancement (CDE) Indigenous Mortality Quality Study. This study linked death records for an 11 month period to the 2006 Census records, which enabled comparison of the reporting of Indigenous status in the two records. More detail on the CDE Indigenous Mortality Quality Study is provided in Chapters 4 and 5 of the Discussion Paper. In summary, the study indicated that the coverage of Indigenous deaths in death registration data was higher than previously estimated (based on the indirect methods).
The outcomes for Indigenous life expectancy for 2005-2007 using direct methods are presented below. The first series (Unadjusted) is based on no adjustment to deaths data, i.e., it is assumed that death registration data are complete and is provided primarily for completeness of analysis. The second series (CDE adjusted) is derived by adjusting the death registration data by the undercoverage observed in the CDE Indigenous Mortality Quality Study.
While the results using the CDE Indigenous Mortality Quality Study do result in some outcomes that are difficult to reconcile (e.g. the results for both WA and NT indicate that death registrations represent an overcoverage of Indigenous deaths relative to Indigenous status in the Census), they do represent the only direct measure of undercoverage of Indigenous deaths available.
ABS PREFERRED APPROACH
The ABS is of the view that a direct method using the CDE Indigenous Mortality Quality Study presents the best option for deriving measures of Indigenous life expectancy for 2006. The approach based on the CDE Indigenous Mortality Quality Study does have limitations. However, it is data based and does not require assumptions. Also, confidence intervals around any error in the estimates can be established. It also produces results that are considered to be more plausible, particularly in regard to the coverage of death registrations.
As indicated above, the indirect methods require extensive assumptions and produce significantly implausible results in terms of outcomes in relation to the apparent undercoverage measures over time. Furthermore, the analysis undertaken has highlighted the sensitivity of the estimates to the accuracy of the population estimates.
The preliminary results for 2006 by state/territory (for those for which estimates can be reasonably compiled) using the preferred approach are presented in the table below.
COMPARISON TO NON-INDIGENOUS LIFE EXPECTANCIES
Table 1.4 presents the difference between Indigenous life expectancy by state/territory (for which data can be compiled) and the Australia level non-Indigenous life expectancy using the CDE Indigenous Mortality Quality Study to adjust Indigenous death registration data.
The results indicate a difference at the Australia level of 11.8 years for males and 10.0 years for females. The estimates vary considerably at the state/territory level with NT (17.2 years for males; 13.4 years for females) being the highest, while NSW (9.1 years for males; 7.8 years for females) was the lowest of the states for which estimates could realistically be compiled. At the same time the standard errors and other limitations (see Chapter 5) associated with these estimates need to be clearly kept in mind.
COMPARISON WITH PREVIOUS PERIODS
The above Indigenous life expectancy estimates are markedly higher than the experimental estimates published for 1996-2001, which were based on an indirect method (Bhat with unexplained growth). In terms of trying to assess changes in life expectancy over time, it needs to be recognised that the above concerns identified in relation to indirect methods are equally applicable to the ABS' previously published Indigenous life expectancy estimates for the period 1996-2001. While the ABS took considerable care to explain the underlying assumptions, and that the estimates were of an experimental nature, it would seem clear now that the derived coverage rates and resulting life expectancy estimates for some states/territories are implausible. The ABS would strongly advise against any comparison of the published Indigenous life expectancies based on 1996-2001 with the estimates presented in this Discussion Paper, and in no way should those differences be interpreted as measuring changes in life expectancy over time.
While the CDE Indigenous Mortality Quality Study does not provide a basis for deriving life tables and life expectancy estimates for previous periods, a range of possible outcomes for previous periods using different assumptions regarding undercoverage of Indigenous deaths in 2000-2002 are presented in Chapter 5 of this Discussion Paper. The results would suggest that the use of indirect methods for 1996-2001 may have resulted in an understatement of Indigenous life expectancy at birth for some states/territories and at the Australian level. For example, if death coverage was 20 percentage points less for 2000-2002 relative to the CDE estimates for 2005-2007, then the life expectancy estimate using a direct adjustment method would be 1.8 years higher for males and 2.4 years higher for females than the previously published ABS estimates (59.4 years for males and 64.8 years for females). Unfortunately, it is not possible to quantify the change in death coverage over the time frame.
The analysis undertaken for 2006 has highlighted the complexities in compiling Indigenous life tables and life expectancy estimates. It is clear that the estimates are subject to a range of potential errors in terms of data inputs, for example, the population estimates are subject to the sampling error associated with the Post Enumeration Survey, and the estimates of undercoverage in death registration data are subject to under or over estimation. In that context it needs to be recognised that life expectancy estimates are just that, estimates based on the available information.
The analysis for 2006 has highlighted the limitations of indirect demographic methods in the compilation of Indigenous life tables and life expectancy estimates. Furthermore, it has emphasised the importance of having direct measures of the extent of undercoverage of death registrations data to provide a basis for adjustment in the compilation of these estimates. The ABS would see such measures as being integral to the compilation of future life tables and life expectancy estimates. The ABS would view a repeat of the CDE Indigenous Mortality Quality Study as critical to the compilation of Indigenous life tables and life expectancy estimates following the 2011 Census of Population and Housing.
Needless to say the difficulties associated with the compilation of life tables and life expectancy estimates are exacerbated when trying to monitor changes over time.
OUTLINE OF SUBSEQUENT CHAPTERS
The remaining chapters of the Discussion Paper cover the following aspects.
Chapter 2 describes the requirements for the compilation of accurate life tables and details the limitations of the population and death estimates for Indigenous Australians, and the need to adopt alternative methods to adjust Indigenous death registrations to account for undercoverage.
Chapter 3 presents discussion of the indirect methods considered in the compilation of life tables and life expectancy estimates. In particular, it discusses the derived coverage rates arising from the methods and the suitability of their use in the Australian context.
Chapter 4 describes a direct demographic method using information from an ABS data linkage project as the basis for adjusting for undercoverage in Indigenous death registrations.
Finally Chapter 5 discusses the outcomes in terms of life expectancy estimates that result from the use of the different methods and presents the ABS preferred method.
ISSUES FOR CONSIDERATION
ABS would welcome comments on any aspects of this Discussion Paper but in particular would appreciate comments as to:
Comments should be forwarded by 15 December 2008 to:
Australian Bureau of Statistics, Locked Bag 10
BELCONNEN ACT 2616
Phone: (02) 6252 6411
In addition, the ABS will be contacting key stakeholders to discuss issues identified in this paper.
These documents will be presented in a new window.