3302.0.55.003 - Experimental Life Tables for Aboriginal and Torres Strait Islander Australians, 2005–2007  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 25/05/2009   
   Page tools: Print Print Page Print all pages in this productPrint All

APPENDIX 2 ALTERNATIVE APPROACHES TO ADJUST DEATHS


INTRODUCTION

The method adopted by the ABS to compile Indigenous life tables has two key features. First, the use of the CDE Indigenous Mortality Quality Study enables direct calculation of identification rates. Second, by aligning the deaths data to the population estimates derived from the 2006 Census and Post Enumeration Survey the methodology assures consistency between the numerator (that is, estimates of deaths) and the denominator (estimates of population at risk).

This appendix presents Indigenous life expectancy estimates using a range of alternative approaches for adjusting deaths data as input to compilation of life tables using the direct demographic method.


ALTERNATIVE APPROACHES TO ADJUST DEATHS

Six alternative approaches were considered:

1. using deaths identified as Indigenous in either the Census or death registrations in the CDE linked data only;

2. using deaths identified as Indigenous in either the Census or death registrations in the CDE linked and unlinked data;

3. using the Chandra Sekar and Deming 'Capture/Recapture' approach;

4. using Indigenous identification based on the 2006 Census;

5. using CDE-adjusted Indigenous deaths, with further adjustments to the age and sex structure of deaths according to preliminary results from the Australian Institute of Health and Welfare Indigenous Mortality Data Linkage Study; and

6. using deaths identified as Indigenous in the Australian Institute of Health and Welfare Indigenous Mortality Data Linkage Study.

The first three approaches were suggested in feedback on the Discussion Paper: Assessment of Methods for Developing Life Tables for Aboriginal and Torres Strait Islander Australians, 2006 (cat. no. 3302.0.55.002). ABS subsequently explored these approaches but considered them to be unsuitable for estimating Indigenous mortality. The main reason for this is that the identification of Indigenous status in the numerator (deaths) and the denominator (population estimates) are inconsistent with each other, thus introducing bias to life expectancy estimates.

The fourth approach was explored by ABS before releasing the discussion paper, but the resultant life expectancy at birth estimates were not published as they were considerably higher than the estimates released in the discussion paper.

The fifth and sixth approaches were explored as a result of discussions with the Australian Institute of Health and Welfare (AIHW) following the release of the discussion paper.

A brief discussion of each of the different approaches and the results obtained are presented below.


1. Deaths identified as Indigenous in either Census or death registrations in the CDE linked data only

ABS explored this option to derive identification rates for Indigenous deaths and life expectancy estimates using the CDE linked data.

As shown in Table A2.1, in the linked file there were 1,327 deaths identified as Indigenous in death registrations in Australia. An additional 323 deaths were identified as Indigenous in the Census only (these deaths were not identified as Indigenous in death registrations). This gives a total of 1,650 Indigenous deaths in the linked file, and an identification rate of 1,327 / 1,650 * 100 = 80% for Australia.

This lowers the identification rate and results in lower life expectancy at birth estimates.

This approach uses a concept of Indigenous status which is different from that used for the denominator for death rates; that is, population estimates where Indigenous status is as reported in PES only but not augmented by other sources of information. In calculating life expectancy estimates, it is important to ensure that the classification of records as Indigenous occurs in a consistent manner in the numerator (deaths) and the denominator (population estimates) as different population scopes feed directly into bias in death rates.

A2.1 CDE Linked data(a), Identification rates and life expectancy estimates based on Indigenous deaths - 2005-2007

Life expectancy at birth estimates
Indigenous deaths in death registrations
Additional Indigenous deaths identified in Census only
Total Indigenous deaths by death registrations and Census
Indigenous deaths identification rate(b)
Males
Females
State/territory
no.
no.
no.
no.
years
years

NSW
372
143
515
0.72
67.5
73.0
Qld
351
62
413
0.85
66.9
72.4
WA
254
34
288
0.88
61.7
67.3
NT
204
5
209
0.98
59.7
67.7
Aust.(c)
1 327
323
1 650
0.80
65.5
71.4

(a) Deaths identified in either Census or registration in the CDE linked data.
(b) Ratio of Indigenous deaths identified in death registrations to the total Indigenous deaths identified in both registration and Census.
(c) Includes all states and territories.



2. Deaths identified as Indigenous in either Census or death registrations in the CDE linked and unlinked data

This approach, like the previous approach, is based on Indigenous deaths identified in either the Census or death registrations, but uses the Indigenous deaths identified in the CDE linked file plus those Indigenous deaths which were unable to be linked to a Census record.

At the national level, there were 1,800 Indigenous death records available from death registrations for linking in the CDE study. This approach uses all of these Indigenous death records in the numerator to estimate the identification rate, as opposed to 1,327 linked records used by the approach above. Of the 1,800 records, there were 1,650 records on the linked file where the deceased was assumed to be Indigenous based on information from their death registration or Census records (table A2.1 above). The 473 (that is, 1,800 minus 1,327) Indigenous death records which were unable to be linked to Census records were then added to the 1,650 Indigenous death records to get the total number of Indigenous deaths available for analysis after linkage. The identification rate was then calculated as the ratio of Indigenous deaths prior to linkage to Indigenous deaths after linkage; that is, 1,800 / 2,123 * 100 = 85% for Australia.

A2.2 CDE Linked and Unlinked data(a), Identification rates and life expectancy estimates based on Indigenous deaths - 2005-2007

Life expectancy at birth estimates
Indigenous deaths prior to linkage
Indigenous deaths after linkage
Indigenous deaths identification rate(b)
Males
Females
State/territory
no.
no.
no.
years
years

NSW
453
594
0.76
68.3
73.6
Qld
429
493
0.87
67.2
72.7
WA
387
422
0.92
62.3
67.9
NT
341
345
0.99
59.9
67.9
Aust.(c)
1 800
2 123
0.85
66.0
71.9

(a) Deaths identified in either Census or registration in the CDE linked and unlinked data.
(b) Ratio of Indigenous deaths prior to linkage to Indigenous deaths after linkage.
(c) Includes all states and territories.


Like the previous approach, this approach also produces life expectancy estimates which are biased as the identification of Indigenous status in the numerator and the denominator are inconsistent.


3. Chandra Sekar and Deming 'Capture/Recapture' model

The ABS has explored the Chandra Sekar and Deming 'Capture/Recapture' approach to deriving identification rates based on the CDE Indigenous Mortality Quality Study data. The logic of this approach is that when a vital registration is incomplete, data obtained from an independent collection can be used in conjunction with the registrar's list to estimate (1) the total number of births and deaths in an area over a specified period, and (2) the completeness of registration. This method was originally applied in a health centre area near Calcutta, India to estimate total numbers of births and deaths and registration completeness for the years 1945 and 1946. The study used data collected from a registration system and an enquiry in the form of a house-to-house canvass (Sekar and Deming, 1949). This approach was also applied to data from a registration system and a survey in conjunction with the 1940 and 1950 Censuses of the United States and the Population Survey in 1969-70 to measure completeness of birth registration (Shryock et al., 1970).

This approach was explored using the CDE linked file of the form:

A2.3 Indigenous status classification

Death registration Classification
Census classification
Indigenous
Non-Indigenous(a)
Total

Census  
Indigenous
a
b
S
Non-Indigenous(a)
c
d
..
Total  
R
..
N

(a) Includes records where Indigenous status was not stated.


where:
  • a = number of deaths recorded as Indigenous in both death registrations and Census;
  • b = number of deaths recorded as Indigenous in Census only, but recorded as non-Indigenous in death registrations;
  • c = number of deaths recorded as Indigenous in death registrations only, but recorded as non-Indigenous in Census, and;
  • d = Indigenous deaths missed by both death registrations and Census.

The values a, b and c are observed and d is estimated as b * c / a on the assumption that the process of contributing to the rows is independent of contributing to the columns. An estimate of the total number of Indigenous deaths can be obtained as:
  • N = a + b + c + d

The identification rate for Indigenous deaths can then be estimated as:
  • identification rate = total number of Indigenous deaths captured in death registrations / estimated total number of Indigenous deaths captured in death registrations and Census
  • = R / N

where R denotes the total number of Indigenous deaths captured in death registrations and S denotes the total number of Indigenous deaths captured in Census.

Identification rates and life expectancy at birth estimates based on this method are presented in table A2.4.

A2.4 Chandra Sekar and Deming 'Capture/Recapture' method, Indigenous deaths identification rates and life expectancy at birth estimates - 2005-2007

Life expectancy at birth
Indigenous deaths captured in registration
Indigenous deaths captured in registrations and census
Identification rate
Males
Females
State/territory
no.
no.
no.
years
years

NSW
372
567
0.66
66.3
71.9
Qld
351
423
0.83
66.6
72.1
WA
254
300
0.85
61.1
66.8
NT
204
209
0.98
59.7
67.7
Aust.(a)
1 327
1 755
0.76
64.6
70.6

(a) Includes all states and territories.


The underlying model is based on the following assumptions:
  • (i) the data collections are independent of each other;
  • (ii) some Indigenous records are classified as non-Indigenous in both data collections;
  • (iii) the population of deaths is divided into 'underlying Indigenous status'; and
  • (iv) non-Indigenous deaths are never reported as Indigenous.

Assumption (i) is valid in this case though it is likely that the same family member could have provided information for some of the linked deaths/Census records. Virtually all deaths that occur in Australia are registered, but not all Indigenous deaths are registered as such. Similarly, some Indigenous deaths are misclassified as non-Indigenous in Census. Therefore, assumption (ii) may be applicable.

The key problem for using this method to derive identification rates for Indigenous deaths based on the CDE linked data relates to the assumption (iv). As can be seen in table 3.3 of Chapter 3: Data linkage to derive Indigenous deaths identification rates, a significant number of deaths were identified as Indigenous in death registrations but were classified as non-Indigenous in the Census, and vice versa. Like the previous two methods, the 'Capture/Recapture' method maximises the number of Indigenous deaths identified by two sources, but disregards the misclassification of Indigenous status reported in Census and registration data. Thus Indigenous classification used in the numerator (deaths) and denominator (population estimates) are again inconsistent.


4. Direct method using identification based on the 2006 Census

The CDE linked dataset was used to derive direct estimates of mortality of the Indigenous population counted in the 2006 Census. The method of estimation was a three stage process.

Firstly, people who identified themselves as Indigenous in the 2006 Census were selected from the Census. There were 455,028 such people, comprising the denominator of age-specific death rates used for life table calculations.

Secondly, only those death records for which Indigenous status was reported as Indigenous in the Census were taken from the linked file. There were 1,379 such records. The CDE study linked 2006 Census records with deaths that occurred from 9 August 2006 to 30 June 2007 for all states/territories (except for Victoria where death records were only available to mid-March 2007). As one full year of deaths are required to derive life tables, deaths were adjusted to account for this part year reference period before using them to compile life tables.

Thirdly, age-specific death rates were calculated by dividing the Indigenous deaths by the Indigenous Census counts. These rates were then used to derive Indigenous life tables. No adjustments were made to the death rates to account for potential undercoverage of deaths, hence the method is referred to as 'direct'.

At the Australia level, this approach produced life expectancy at birth estimates of 71.9 years for Indigenous males and 76.4 years for Indigenous females. These estimates are much higher than those derived after adjusting Indigenous deaths by the undercoverage figures obtained from the CDE study. No life tables were produced at the state/territory level as the number of deaths involved were small.

This approach was intended to make Indigenous counts for deaths and for the population at risk (that is, Indigenous usual residents of Australians) consistent. However, this is not the case. Because of undercount in Census a proportion of death records could not be linked, as there was no corresponding Census record to link with. Such people are removed from the counts of deaths and from the counts of the population at risk. However, people whose information from death registrations or Census is not accurate enough to make a link are not included in the death counts, but it is not possible to remove from the Census count the corresponding count of people with unlinkable information in Census or what would have been unlinkable information from death registrations, had they died.


5. CDE-adjusted Indigenous deaths with further adjustments to age structure of deaths according to AIHW Indigenous Mortality Data Linkage Study

Following the release of the discussion paper, the Australian Institute of Health and Welfare provided preliminary results from the Indigenous Mortality Data Linkage Study to ABS. The study has linked deaths from the AIHW National Mortality Database to the AIHW National Hospital Morbidity Database, the AIHW National Perinatal Data Collection, the Aboriginal Health Liaison Officer Dataset and to the Australian Government Department of Health and Ageing Residential Aged Care Dataset for 2001-2006 to enhance identification of Indigenous deaths.

From these data, estimates of Indigenous life expectancy at birth taking into account differential identification of Indigenous deaths according to age and sex were derived.

ABS calculated age-specific Indigenous deaths identification rates for Australia directly from the AIHW data, which were applied to the number of CDE-adjusted deaths (these are the numbers of Indigenous deaths used to produce the ABS estimates in this publication) to obtain age-adjusted numbers of Indigenous deaths. These deaths were then used as numerators in the calculation of abridged life tables.

For Australia, this approach resulted in life expectancy at birth estimates of 67.1 years for Indigenous males and 72.7 years for Indigenous females (table A2.5).

It should be noted that the identification rates from the AIHW study relate to the period 2001-2006 while the number of CDE-adjusted deaths relate to 2005-2007. Nevertheless, the estimates derived from this approach are very similar to the ABS estimates for 2005-2007 (67.2 years for Indigenous males and 72.9 years for Indigenous females; table 1.1).

A2.5 Life expectancy at birth of Indigenous Australians, Incorporating results of the AIHW Indigenous Mortality Data Linkage Study - 2005-2007

Life expectancy at birth
Males
Females
years
years

CDE-adjusted deaths with AIHW age adjustments
67.1
72.7




6. Direct method using deaths identified as Indigenous in AIHW Indigenous Mortality Data Linkage Study

The AIHW provided estimates of Indigenous life expectancy at birth based on complete life tables using the average yearly number of deaths for 2001-2006 identified as Indigenous in the Indigenous Mortality Data Linkage Study. For Australia, this resulted in life expectancy at birth estimates of 67.1 years for Indigenous males and 73.1 years for Indigenous females (table A2.6).

ABS also calculated Indigenous life expectancy at birth from these data, using abridged life tables, consistent with the methodology used to produce the ABS life expectancy estimates presented in this publication (abridged life tables were chosen as age-specific death rates for 5-year age groups were considered more reliable than those for single years of age due to the small annual numbers of Indigenous deaths in the states and territories). For Australia, this approach resulted in life expectancy at birth estimates of 66.7 years for Indigenous males and 72.6 years for Indigenous females (table A2.6). The small differences between these and the AIHW estimates derived from complete life tables may be due to the choice of method for approximating mortality for 85 years and over.

Both of the above estimates derived from the AIHW Indigenous Mortality Data Linkage Study are consistent with ABS Indigenous life expectancy estimates presented in this publication.

A2.6 LIFE EXPECTANCY AT BIRTH OF INDIGENOUS AUSTRALIANS(a), AIHW Indigenous Mortality Data Linkage Study - 2001-2006

LIFE EXPECTANCY AT BIRTH
95% CONFIDENCE INTERVALS
Males
Females
Males
Females
years
years
years
years

Complete life table(b)
67.1
73.1
66.7-67.4
72.7-73.5
Abridged life table(c)
66.7
72.6
..
..

(a) Direct method using deaths identified as Indigenous in AIHW Indigenous Mortality Data Linkage Study, and an average of 30 June 2001 and 30 June 2006 Experimental Estimates of Aboriginal and Torres Strait Islander Australians.
(b) Calculated by AIHW.
(c) Calculated by ABS.



References

Sekar, Chandra and W. Edwards Deming (1949), On a method of estimating birth and death rates and the extent of registration, Journal of American Statistical Association, vol. 44, No. 245, pp. 101-115.

Shryock H, Siegel J and Associates (1976), The Methods and Materials of Demography, Condensed Edition, Academic Press, New York.