The 2011 census population counts can be decomposed by sex, age, current residence, residence five years ago, and current identification status. In addition, the ACLD links a sample of the 2006 and 2011 censuses.

The ACLD was released by the ABS in late 2013, and ‘a sample of almost one million records from the 2006 Census (wave 1) was brought together with corresponding records from the 2011 Census (wave 2) to form the largest longitudinal dataset in Australia’ (ABS 2013c). To produce the ACLD, five per cent of records from the 2006 census were linked probabilistically with available data from the 2011 census based on the most likely match, given observed characteristics. Because this linking was done without knowing the individual’s exact name and address, an unobserved minority of linked pairs are not of the same individual. We use data from the ACLD to calculate the age-specific probabilities of identification change between the two census periods.

Mortality data are sourced from the 2010–12 Indigenous life tables (ABS 2013d), while fertility rates are sourced from vital registration data compiled in the ABS (2015) Births publication. Net international migration data for the 2006–2011 period for the non-Indigenous population data were obtained from the Australian Bureau of Statistics, and represent the difference between immigration and emigration. International migration is measured using registration (passenger card) data. An immigrant (emigrant) is defined as a person who has entered and remained in (left and remained outside of) Australia for 12 out of 16 months. Finally, for the projections, the 2011 Estimated Resident Population is used as the starting point.

The population counts of the 2011 census by usual residence in 2006 (t), usual residence in 2011 (t+5), and identification status in 2011 are presented in Table 1.² The major city, regional area, and remote area definitions are taken from the ABS (2013b). Collectively, they comprise a geographic area classification known as ASGS-RA, which reflects the relative proximity of every location in Australia to towns and cities. The classification is created by constructing a one-kilometre grid for all of Australia and assigning an index value, known as the Accessibility or Remoteness Index of Australia (ARIA+), to each square in the grid based on the average road distance from the square to the nearest town or urban centre of various population sizes. The grid is laid on top of the Statistical Area 1 (SA1) geographic structure to produce an ARIA+ score for each SA1. SA1s with an ARIA+ score of 0.0 to 0.2 are classified as major cities; of 0.2 to 5.9, as regional areas; and of greater than 5.9, as remote areas. These area classifications are presented for Australia in Figure 1. For example, an SA1 that lies within an urban area with a population greater than 250,000 or within one-fifth of the average distance to such a centre is classified as major city. An SA1 that is 0.5 times the average distance to a major city, 1.2 times the average distance to a town with between 48,000 and 250,000 people, and sits inside a town with between 18,000 and 48,000 people obtains an ARIA+ score of 1.7 (0.5+1.2+0.0), and is classified as a regional area (0.2<1.7≤5.9). Meanwhile, remote areas are farthest from major urban centres. More information on how ARIA+ is calculated is available from the Hugo Centre for Migration and Population Research at the University of Adelaide.³

Table 1 shows that of the total Indigenous population in 2011, 185,000 (43 per cent) lived in regional areas, 150,000 (35 per cent) lived in major cities, and 95,000 (22 per cent) lived in remote areas. In contrast, of the total non-Indigenous population in 2011, 11.8 million (70 per cent) lived in major cities, while only 277,000 (2 per cent) lived in remote areas. The shares of the populations living in these area groups did not change much between 2006 and 2011, but the population numbers grew by 11.9 per cent in major cities, 12.6 per cent in regional areas, and 8.2 per cent in remote areas.

Looking at the data on inter-area group migration in Table 1, we see some notable differences between the Indigenous and non-Indigenous populations. First, the Indigenous people living in major cities were more likely to move to a regional area or remote area (i.e., 9,133+1,093/147,335=7 per cent) than the non-Indigenous (i.e., 445,045+35,239/11,812,917=4 per cent). The opposite was the case for those living in regional and remote areas, where the Indigenous population had lower out-migration propensities than the non-Indigenous population. For example, almost 30 per cent of the non-Indigenous population but only 10 per cent of the Indigenous population living in remote areas moved to a major city or regional area between 2006 and 2011. Both populations exhibited higher propensities to move to regional areas when leaving major cities or remote areas. When moving from regional areas, both groups preferred to move to major cities.

The following multistate demographic accounting model (Rogers 1995) is employed to project the Indigenous and non-Indigenous populations of Australia.

K(t+5)=GK(t+5)+I  (1) where K(t)=[K0(t)K5(t)⋮K75+(t) ],Kx(t)=[K11x(t)K12x(t)⋮K23x(t) ],Ix=[I11xt,t+5I12xt,t+5⋮I23xt,t+5 ], and G=[000B10⋯B450⋯0S0⋱0S5⋱⋮⋮⋮⋱000⋯S700].

Kiax(t) is the population at time t by identification status i (1= Indigenous, 2= non-Indigenous), by area group a (1= major cities, 2= regional areas, 3= remote areas), and by age group x (0−4,  5−9,…, 75+). Iiax reflects the corresponding net immigration totals between the two time points. As specified in Rogers (1995), Sx and Bx are six-by-six matrices that are composed of the survival probability of moving from identification status i and area group a to identification status j and area group b; and the survival fertility rate of women in identification status i and area group a contributing births to women in identification status j and area group b, respectively. The projection is performed separately for males and females. The six subpopulations are defined as Indigenous major cities (K11), Indigenous regional areas (K12), Indigenous remote areas (K13), non-Indigenous major cities (K21), non-Indigenous regional areas (K22), and non-Indigenous remote areas (K23).

In our model framework, births are produced by females. In the construction of the fertility rates, we assume that women can give birth to both Indigenous and non-Indigenous babies. Non-Indigenous women give birth to Indigenous children largely by partnering with Indigenous men (Biddle and Johnstone 2014); whereas a much smaller percentage of Indigenous women give birth to non-Indigenous children. While this pattern may seem counter-intuitive, it likely reflects personal factors that influence how people identify themselves and their children.

In this paper, we extend the multistate population projection of O’Donnell and Raymer (2015), which included age-specific transitions between Indigenous and non-Indigenous identification, by including transitions between residence area groups. Many standard cohort-component population projections models ignore transitions between states of residence, and instead rely on net migration or the slightly more detailed out-migration and in-migration rates. The problem with net migration and in-migration rates is that they include the incorrect population at risk of migrating in the denominator, which can seriously bias the projection results (Rogers 1990, 2015). The same is true for other life course transitions in which populations can move in both directions; e.g., ethnic self-identification, marriage, health, and employment.

In order to account for the identification status transition and migration among area groups simultaneously, we need to estimate the rates Sx described in Equation 1. Although such information can be extracted from the ACLD directly, the quality of this approach is questionable. For one thing, the ACLD only accounts for five per cent of the total Australian population data. Thus, after the decomposition, many of the numbers in the cells of the above matrix are small or zero, especially when disaggregated by age and sex. Furthermore, small counts have been randomized by the ABS to protect individual confidentiality. Therefore, we propose combining the ACLD data with the 2011 census data to produce more reliable inputs for analysis and use in a multistate population projection.

The aggregated smoothed census counts grouped by the usual residence and identification status in 2006 (t) and in 2011 (t+5) are presented in Table 3. As the identification statuses are estimated by combining data from the ACLD data and 2011 census, the observations in Table 3 and Table 1 are consistent in the 2011 (column sum) margins with some minor differences (less than 0.01 per cent) due to the smoothing and the application of the estimated age-specific probabilities.

From the two off-diagonal panels of Table 3, we can see that only 0.27 per cent (i.e., (23,537+19,935+1980)/16,828,654) of the total non-Indigenous population transitioned to Indigenous status during the 2006–2011 period, compared with 6.80 per cent (i.e., (14,002+12,725+1243)/411,596) for the Indigenous population. For the Indigenous people who were living in the three area groups and changed their identification to non-Indigenous, the percentages were 10.15 per cent (i.e., (13,208+644+119)/137,579 for major cities, 7.17 per cent (619+11,944+30)/175,673) for regional areas, and 1.43 per cent (175+137+1094)/98,345) for remote areas. The corresponding percentages for the non-Indigenous people transitioning to Indigenous status were 0.20 per cent, 0.42 per cent and 0.77 per cent, respectively. In terms of the population counts, there were 26,245 Indigenous individuals who changed their identification status to non-Indigenous, but stayed in their area group of residence. The number of Indigenous people who changed their identification and moved to a different area group was 1725, or about 6.2 per cent of this group. For non-Indigenous people, the corresponding numbers were 42,953 who changed their identification but remained, and 2498 (5.5 per cent) who changed their identification and moved.

Both identification change and internal migration contributed to population change in each of the three areas. Major cities gained 9535 Indigenous people through identification change (i.e., 22,354-13,208), and another 2509 people through internal migration (i.e., 9816+1903-8208-1003). Regional areas also experienced net gains in Indigenous population through both identification change (18,768-11,944=6824 persons) and internal migration (8208+7084-9816-3736=1740 persons). However, the Indigenous population in remote areas declined because of a large negative net internal migration of 4248 people, which was substantially more than the 737 individuals gained through identification change.

As suggested by Figures 2 to 5, and as shown in previous individual-level modelling (Biddle and Crawford 2015), the interaction between inter-area group migration and identification status change is significant. To further explore this interaction, we present the ratio of the odds of people changing their identification and migrating to the odds of people changing identification (but not migrating) in Table 4. For example, consider the odds ratio of 7.41 for those changing their status from Indigenous and living in remote areas to non-Indigenous and living in major cities. This odds ratio is calculated from the counts presented in Table 3 and represents: the ratio of (i) the odds of moving from remote areas to major cities and changing identification to non-Indigenous moving from remote areas to major cities and not changing identification (i.e., 175/1903=0.092) to (ii) the odds of changing identification to non-Indigenous to those remaining Indigenous (i.e., 1094/87,951=0.012). These types of moves clearly had strong and important effects on identification change.

Table 4 contains very useful information regarding the relationships between migration between different area groups and identification change. Like Indigenous people who were moving from remote areas to major cities, Indigenous people who were moving from remote areas to regional areas were more likely than those who were not moving to change their identification, albeit with a smaller odds ratio of 1.56. However, Indigenous people who were undertaking most of the other kinds of moves retained their status at much higher rates than those who did not move. This was particularly true for Indigenous people who were moving from regional areas to remote areas, for whom the odds ratio was 0.10.

Among the non-Indigenous population, those who moved from major cities to regional areas and remote areas exhibited odds ratios of 1.13 and 1.55, respectively; indicating positive relationships between these moves and identification change. The other moves had the opposite effect: i.e., the non-Indigenous people who moved from regional areas and remote areas were much less likely to change their identification to Indigenous.

Assuming fixed values for fertility, mortality, and inter-area group migration components, the 2011–2031 projected populations for the three areas and the corresponding sources of growth are presented in Table 5. Comparing the 2011 Estimated Resident Populations and the 2016 projection, we see that the Indigenous population was projected to have increased by 37,730 (16.2 per cent) in major cities, by 40,777 (13.9 per cent) in regional areas, and by 7792 (5.5 per cent) in remote areas. The most important projected source of growth was natural increase (i.e., births - deaths). For major cities and regional areas, the second-most important projected source of growth was net identification change. For remote areas, however, the second-most important source of growth was (negative) net internal migration.

Over two-thirds of the negative net migration of Indigenous persons in remote areas was added to major cities. Identification change contributed positively to Indigenous population growth in all three areas, but especially in major cities. The contributions of people changing both their identification status and their area of residence represented only a small share of the overall population change.

Comparing the projected sources of growth during the 2016–2021, 2021–2026, and 2026–2031 periods, we see that most of the above observations hold, albeit with some noticeable differences. First, levels of natural increase grow steadily over time, from 28,000 to 40,000 for major cities, from 36,000 to 44,000 for regional areas, and from 12,000 to 13,000 for remote areas. Second, levels of net internal migration decrease over time for major cities and remote areas, but increase slightly for regional areas. Third, net identification change to Indigenous status is positive across all three area groups and projection periods, but the levels fluctuate for major cities, decline for regional areas, and increase for remote areas. The net migration and identification changes are relatively small and contribute very little to overall growth, albeit with consistent negative growth for major cities and remote areas and positive growth for regional areas. Overall, we expect Indigenous populations to continue growing in all three area groups in Australia. Between the 2011–2016 and 2026–2031 periods, growth is projected to increase by around 32 per cent for major cities (from 38,000 to 50,000) and remote areas (from 8000 to 10,000). For regional areas, growth is projected to increase by 18 per cent.

We now proceed to describe the multistate projections to 2031 holding the 2006–2011 rates constant. The age-sex compositions of the 2011 Estimated Resident Population (dark shaded bars) and projected 2031 population (diagonal shaded bars) for the Indigenous population are presented for the three areas in Figure 8. There are two conclusions we can draw from these percentage plots and the corresponding plots of population counts (not shown). First, consistent with our observation in Section 4.1, growth is occurring in all area groups and age groups, but particularly in major cities and regional areas. Second, ageing is occurring faster in major cities and regional areas, even though the ‘pyramid’ shape remains more or less the same.

We demonstrated in Section 4.1 that identification change contributes greatly to Indigenous population growth in all three areas. To further verify this point, we compare in Figure 9 three projections of the Indigenous population using our multistate model and a single state model that includes no identification change and fixed net internal migration levels for the years 2011 to 2031. We find that the multistate projection results in larger Indigenous populations in all areas due to the inclusion of identification change. The corresponding ABS projections (not shown) were between the multistate and single state models for major cities, close to the multistate model for regional areas, and below the single state projection for remote areas.