1The Australian Aboriginal and Torres Strait Islander (Indigenous) population comprised approximately 2.8% of the Australian population counted in the 2011 census and around 3.0% of the projected population as of June 2015 (ABS, 2013g; ABS, 2015). The 2011 census enumerated 495,755 people who identified as Aboriginal only; 31,407 as Torres Strait Islander only; and 21,205 as both Aboriginal and Torres Strait Islander (see Appendix Table A.1). There were a further 1,058,586 records for whom Indigenous status was not reported, of which some were allocated to the Indigenous population estimate, based on results from the Post Enumeration Survey (PES). This survey identifies the probability of a census record with no Indigenous status available (due to unit or item non-response) being Indigenous. These probabilities are then applied to the records with Indigenous status missing to obtain a point-in-time estimate of the total Indigenous population.

2Indigenous Australians are of particular policy importance to government for a number of reasons, many of which are shared by other Indigenous groups in settler colonial countries. Geographically, though the majority of the Australian Indigenous population live in urban areas, this population is more heavily concentrated in rural and remote areas than their non-Indigenous counterparts. Demographically, they are concentrated in younger age groups, in particular those of preschool and school age as well as those transitioning between school and further education and/or employment. Indigenous Australians are more likely than non-Indigenous Australians to suffer economic hardship and have lower rates of employment across the life course, and less likely to own or be purchasing their own home (Biddle and Yap, 2010). There remains a large gap in the life expectancy of Indigenous and non-Indigenous Australians (11.5 years for males and 9.7 years for females; ABS, 2013f). Designing and delivering services to redress the inequality between Indigenous and non-Indigenous people first require a detailed understanding of Indigenous populations.

3The enumerated Indigenous population grew 20.5% in the 2006–2011 census interval. The Indigenous Estimated Resident Population (ERP) grew at an even higher rate, 29.6%, which may be attributed to improvements in the Australian Bureau of Statistics’ methodology in estimating census undercount (ABS, 2013b). The observed population growth in this period was higher than that in both the 2001–2006 (11%) and 1996–2001 (16%) intervals but consistent with growth observed historically. Averaging 4% per annum over the past 40 years, this population growth is much higher than that of the non-Indigenous population.

4Several factors are contributing to this rapid growth. First, there is natural increase – more Indigenous births than deaths in a given period. In 2013, the number of childen ever born for Indigenous women was 2.34 births per woman compared with 1.88 for all women (ABS, 2014). High rates of both male and female exogamy are likely to further increase the number of births identified as Indigenous. Also contributing to natural population growth is changing Indigenous mortality. Despite long-standing data-quality issues, notably misidentification of Indigenous status on death records, the data suggest that Indigenous life expectancy is slowly increasing (ABS, 2013a). Higher fertility rates, exogamy, and life expectancy all contribute to natural population increase.

5The second factor is improved enumeration. Compared to the non-Indigenous population, the Indigenous population is concentrated in remote areas and is more mobile on a temporary basis (Taylor, 1997), making it difficult to obtain accurate population counts. Improved enumeration has been noted as a key factor in observed Indigenous population growth in historical censuses (see, for example, Dunn, 1992). The ABS refined their Indigenous enumeration strategy for the 2011 census, devoting more staff to undertake fieldwork and collaborating with Indigenous communities prior to and throughout the collection period (ABS, 2012).

6The final factor contributing to population growth is identification. In Australia, a person is Indigenous if they identify as Indigenous, have Indigenous ancestry, and are accepted as Indigenous by their community, though typically only the first component is captured by data collectors (ALRC, 2003). Although the definition appears to imply a certain degree of stability over the life course, Indigenous status can change; for example, a person may discover or reconnect with their Indigenous ancestry. Alternatively, while Indigenous identity may remain constant, there are external factors, such as trust in the ABS or government more generally, that may influence a person’s willingness to identify as such to data collectors (ABS, 2013e).

7Changes in identification have long been noted as a driver of population change. Altman (1992) postulated that identification change may have accounted for up to half the growth in the Indigenous population between the 1971 and 1991 censuses, and its impact on population estimates is being seriously considered in Australia (see Zhang, 2014). Further, identification change is a barrier not just to producing accurate population estimates but to inferring change from analysis of two or more datasets. If a person identifies as Indigenous in one dataset but not another, and if their characteristics are different from those who consistently identify as Indigenous, then their presence and absence in different datasets will result in artefactual differences in outcomes between the measured Indigenous population across the two datasets. However, changes in identification have so far been impossible to study quantitatively in Australia using individual-level microdata because no survey or data source has recorded an individual’s Indigenous status at two or more points in time alongside other explanatory information.

8In this paper, we analyse a relatively new data source, the Australian Census Longitudinal Dataset (ACLD), which consists of a 5% sample of the 2006 census linked to the 2011 census and thus contains a large longitudinal sample of Indigenous Australians over two time points (the largest in Australia). After introducing the ACLD, we analyse and contrast four groups of people: those who were Indigenous in both 2006 and 2011 (Always Identified), those who were Indigenous in 2006 but not 2011 (Formerly Identified), those who were non-Indigenous in 2006 but Indigenous in 2011 (Newly Identified), and those who were non-Indigenous in both 2006 and 2011 (Never Identified). We then assess and contrast these four groups’ characteristics, including age, sex, parents’ Indigenous status, geography, and socioeconomic status. Finally, we assess changing Indigenous status by developing binary regression models predicting transitions into the Indigenous and into the non-Indigenous groups. Throughout the paper, we reflect on the ACLD’s suitability for understanding Indigenous population change and on the role of the linkage methodology in introducing biases into the estimation.

I – The Australian Census Longitudinal Dataset

9Created by the Australian Bureau of Statistics, the ACLD consists of a 5% random sample of the 2006 census (979,662 records) linked to the 2011 census. The target population represented by this file is those in Australia at the two time points. Records on the two files were linked based on fields common to the datasets, notably date of birth, meshblock (a fine-level of geography, where meshblock-5-years-ago is taken from the 2011 census, to align with 2006 residence), sex, Indigenous status, country of birth, year of arrival, along with potentially transitory fields such as marital status, schooling and qualification, and occupation. The census question on Indigenous status, as well as questions on all key explanatory variables, were exactly the same in 2006 and 2011. It is worth noting that there was a reduction in Not Stated Indigenous status between the 2006 and 2011 censuses (Appendix Table A.1), which may indicate, among other things, an increased willingness of some Indigenous people to identify. [1]

10The ACLD represents the population of Australians enumerated in both the 2006 and 2011 censuses. It does not capture population changes due to fertility, mortality, improved enumeration, or international migration, though the latter is a negligible component of Indigenous population change. However, the ACLD is uniquely suited to study changes in Indigenous identification. It is the first longitudinal dataset in Australia with multiple measures of Indigenous status.

11The linkage methodology broadly followed that developed by Fellegi and Sunter (1969), in which each field contributes an agreement (positive) or disagreement (negative) weight of magnitude in proportion to its reliability and discriminatory power. Records may be linked despite disagreement on linking variables, providing there is sufficient agreement on other fields to offset it. [2]

12The sample file contained 979,662 records, of which 800,758 were linked to a 2011 census record. The file contained 21,985 Indigenous persons, of whom 14,806 (67.3%) were linked to a 2011 record. The linked file contains proportionally fewer Indigenous persons than the census, which suggests the Indigenous population was more difficult to link. This difficulty may be attributed to such factors as higher rates of mobility and proportionally more children in the Indigenous population (children being too young to have meaningful values for fields such as qualification, education, marital status, etc.). However, the ABS provides population weights to correct for the under-representation of Indigenous persons on the linked file. The ABS calculated weights with three purposes: first, to weight the linked sample file to an ERP of persons in scope in both 2006 and 2011; second, to correct for biases introduced by the linking process; and third, to benchmark to known subpopulation totals. A full description is provided by the ABS (2013c). We used these weights throughout our analysis. [3]

II – Linking errors in the ACLD

13Two types of linking errors can arise from the linking process: omitting to link a record that ought to be linked (missed records) and linking one individual to a different person with similar characteristics (false links). Of the 979,662 records in the sample, 800,758 were linked to a 2011 census record. The ABS estimates that this equates to a 90% link rate once deaths and emigration are accounted for. Of the linked records, the ABS estimates 90%–95% are linked correctly, while 5%–10% are linked to a different individual with similar characteristics (ABS, 2013d). The potential impact of missed records in the ACLD is that records may be less likely to be linked if they disagree on Indigenous status. Of the 11 linking passes that produced links, Indigenous status was used as a blocking field in six passes (agreement was a necessary condition for records to be linked) and as a linking field in a further four passes (agreement on Indigenous status contributed to deciding whether records were linked. See ABS 2013d for full details). However, the weights assigned to agreement and disagreement on Indigenous status were small compared with other linking fields (such as date of birth and geography). Thus, disagreement on Indigenous status was easily offset by agreement on other fields in those four passes that featured Indigenous status as a linking field. In the remaining pass, Indigenous status was not used in the linking process.

14The bias potentially introduced by the linking process might under-represent the number of Indigenous transitions in the population because records were slightly more likely to be linked if they agreed on Indigenous status. However, there were considerable opportunities for disagreeing records to be linked, and we believe the bias to be small, if at all present.

15False links on the file may have the opposite impact. If we assume that the file contains 10% false links (or roughly 80,000 records) and that these occur at random and are linked to random records, then given that roughly 2.5% of the 2006 population census was Indigenous, we would expect around 78,000 (97.5%) non-Indigenous (in 2006) false links and 2,000 Indigenous. In 2011, the Indigenous population comprised 2.8% of the census, so if these 78,000 non-Indigenous records were linked to a random record, we would expect 75,816 (97.2%) to link to another non-Indigenous record, while 2,184 (2.8%) to link to an Indigenous record. Of 2,000 falsely linked Indigenous (in 2006) records, we would expect 1,944 (97.2%) to link to a non-Indigenous record and 66 to falsely link to another Indigenous person. Under these assumptions, the transitions due to false links come close to cancelling one another out. We would expect 240 (2,184–1,944) more false non-Indigenous to Indigenous transitions than Indigenous to non-Indigenous. This presents the worst-case scenario for our analysis. If the false link rate is 5% rather than 10%, the problem is considerably reduced. It is unclear, however, whether false links occur at random or at higher rates in certain demographic groups. False links are not linked randomly, but they are linked to a record with similar characteristics. Because Indigenous status played a small role in the ABS’s matching process, there may be proportionally fewer false links on the file which change Indigenous status.

16As we did not have direct access to individual records in the ACLD or to detailed information about how each record was linked, we cannot determine the likelihood that a given record was falsely linked. However, we may suspect proportionally more false links in records which change over time on many variables. We consider this throughout our analysis and note in our results a possible small bias towards records being linked when Indigenous status matches across the two time points. Some of the observed transitions will be false links and lead to some noise around our observations and modelling.

III – Changing Indigenous status

17Although a lack of national-level longitudinal data has limited research into changing Indigenous status, studies have demonstrated ethnic mobility (a change over time in how one reports his or her ethnicity) in both Australian and overseas populations. Liebler et al. (2014) used linked US census data (consisting of records from the 2000 US census linked to those from 2010) to investigate changing ethnicity. They estimated 8.3% of the US population changed how they reported their racial identity. They also found differences in the likelihood of change for different racial groups, with Hispanic Americans more likely to change their racial identity than single race non-Hispanic whites, blacks, or Asians. However, Liebler et al. found that transitions primarily amounted to churn and did not contribute to a change in the population of any given ethnic group.

18In a study with similar aims, Carter et al. (2009) analysed characteristics associated with changes in self-identified ethnicity in three waves of survey data in New Zealand (conducted annually). The strongest predictor of change in ethnicity between the second and third waves was ethnicity in the first. Those reporting more than one ethnicity were far more likely to change subsequently; however, those identifying solely as Maori were also more likely to change in later waves.

19In Australia, several studies have used linked health data both to assess changes and to improve recording of Indigenous status. Draper et al. (2009) observed that the increased life expectancy for Indigenous males aligned with increased rates of missing Indigenous status. When Indigenous status was derived from linked records, the observed increase in life expectancy was smaller than when taken from a single source, though still significant. Results varied slightly depending on how the linked records were combined to derive Indigenous status – specifically, whether being Indigenous on any record was sufficient to classify as Indigenous, or more conservatively, whether being Indigenous on the majority of linked records was required. The importance and difficulty of combining multiple and conflicting Indigenous statuses from multiple data sources has been explored. Neville et al. (2011; cited in Thompson et al., 2012) assessed algorithms to better define Indigenous status on mortality files using auxiliary data sources, suggesting that such techniques can reduce misclassification particularly for Indigenous persons living in urban areas. While useful, these administrative datasets have little information on the characteristics of those whose Indigenous identification changed.

20Changing or misreported Indigenous status on the census has been previously analysed using the Post Enumeration Survey (PES). The PES is conducted in Australia several weeks after each census and is used to estimate census undercount (and overcount). As PES data are linked to census data, it provides a data source with more than one Indigenous status. The PES Indigenous population is relatively small, however, and not representative of the Indigenous population as a whole (historically, it was not enumerated in remote communities). Hunter (1998) used census–PES files from both 1991 and 1996 and found that in 1996 there were proportionally fewer transitions from non-Indigenous (census) to Indigenous (PES) than in 1991, suggesting identification rates may have been higher in the 1996 census.

21Changes in Indigenous status and in ethnicity more broadly are therefore anticipated, and their potential impact on analysis (such as mortality estimates) has been demonstrated. However, the changes remain difficult to predict, and factors motivating such changes are not well known. In 2012, the ABS conducted focus-group testing, aiming to better understand the factors affecting Indigenous Australians’ propensity to identify as Indigenous in surveys and the census (ABS, 2013e). The ABS found that factors encouraging identification included personal reasons, such as a sense of pride in one’s identity, along with perceived benefits for themselves or their community. Factors discouraging identification included an expectation that it may lead to racism, discrimination, or other negative repercussions. In some contexts, respondents felt the question was unnecessary and inappropriate, and reported that such circumstances also discourage identification. Demographically, younger people were more likely to self-identify as Indigenous.

2 – Age and sex distribution

23Figure 1 gives the age (in 2011) and sex distribution of the Always, Newly, Never, and Formerly Identified as Indigenous groups. It shows that most transitions (in both directions) occur in children. The populations of the Newly and Formerly Identified groups are concentrated in children more heavily than the Always Identified group. Children do not self-identify; rather, their parents or guardians identify on their behalf. At some point in young adulthood, they begin to self-identify, but we can assume that the 15–19-year-olds and younger (in 2011) did not self-identify in 2006. Many of the observed transitions may be simply due to a different person reporting the child’s Indigenous status in the two censuses, particularly when a child has an Indigenous and non-Indigenous parent.

24The net flows into and out of the Indigenous population are positive across all age-by-sex groups. For both males and females at every point on the age distribution, there are more Newly Identified than Formerly Identified.

Figure 1

Age and sex distribution of the four Indigenous classifications

Age and sex distribution of the four Indigenous classifications

3 – Family consistency

25The ACLD is a sample of persons, not families or households. Family members of those linked do not appear on the linked file unless they were coincidentally in the 2006 ACLD sample. However, some family information is available on person records. A particularly relevant source of information is parents’ Indigenous status for children on the ACLD sample file. This section considers consistency between children’s Indigenous status and that of their parents. It leaves aside the issue of exogamy – children born to an Indigenous and a non-Indigenous parent – and instead attempts to assess whether a change in a child’s Indigenous status mirrors the change in that of either parent.

26Table 1 gives mothers’ Indigenous status in 2006 and 2011 for the Always, Formerly, Newly, and Never Identified children (whatever their age) living with their mother. Generally, changes in children’s Indigenous status track closely with transitions in their mother’s status. The key exception is the group of Newly Identified children born to Never Identified mothers, which is most likely explained by male exogamy (see section below). Some categories should be unpopulated, or at least very rare. Never Identified children born to Always Identified mothers, for instance, is an unlikely event and probably reflects inconsistencies in reporting or the effect of false links in the sample file.

Table 1

Indigenous status of children and their mothers in 2006 and 2011

Child’s Indigenous status 2006 Child’s Indigenous status 2011 Mother’s Indigenous status 2006 Mother’s Indigenous status 2011 Indigenous Non-Indigenous Indigenous Indigenous (Always Identified) Indigenous 119,249 (96.6%) 4,161 (3.4%) Non-Indigenous 4,233 (7.8%) 49,801 (92.2%) Non-Indigenous (Formerly Identified) Indigenous 363 (3.5%) 9,933 (96.5%) Non-Indigenous 0 (0.0%) 4,517 (100.0%) Non- Indigenous Indigenous (Always Identified) Indigenous 999 (100.0%) 0 (0.0%) Non-Indigenous 13,023 (52.9%) 11,611 (47.1%) Non-Indigenous (Formerly Identified) Indigenous 638 (31.8%) 1,366 (68.2%) Non-Indigenous 2,320 (0.1%) 3,831,388 (99.9%)

Indigenous status of children and their mothers in 2006 and 2011

Interpretation: Of the Always Identified children, 119,249 (96.6%) had an Always Identified mother.
Notes: Table pertains to children living with their mothers in 2006, whatever their age. Figures in bold indicate that a mother’s Indigenous status across the two time points matches the Indigenous status of her child.

27Table 2 gives population estimates of the Indigenous status in 2006 and 2011 of children aged 5 to 19 (in 2011) by both parents’ Indigenous status at the same time points. Unsurprisingly, children’s Indigenous status predominately aligns with that of their parents. A child’s transition from Indigenous to non-Indigenous is often also seen in their parents. The converse transition (from non-Indigenous to Indigenous) is less consistently reflected in parental transitions, with this group more likely to have at least one Never Identified parent. There are three likely factors driving the new identification in children. First, that they are children of an Indigenous and non-Indigenous parent, and transitions may be attributed to the person reporting the child’s status. Second, that the same parent is filling out the form, but the parent’s view of the child’s status has changed due to changing social or other circumstances. Third, that their family as a whole has changed status and identified as Indigenous in 2011. As Newly Identified children have both Newly and Never Identified parents, it seems that all three factors contribute to this identification change.

Table 2

Parents’ Indigenous status for 5–19-year-olds in 2011 (%*)

Mother’s Indigenous status 2006 Indigenous status 2006 Indigenous status 2011 Mother’s Indigenous status 2006 Indigenous Non-Indigenous Mother’s Indigenous status 2011 Indigenous Non-Indigenous Indigenous Non-Indigenous Indigenous Indigenous 96.6 3.4 8.3 91.7 Non-Indigenous 2.6 97.4 0.0 100.0 Non- Indigenous Indigenous 100.0 0.0 53.5 46.5 Non-Indigenous 38.4 61.6 0.1 99.9 Indigenous status 2006 Indigenous status 2011 Father’s Indigenous status 2006 Indigenous Non-Indigenous Father’s Indigenous status 2011 Indigenous Non-Indigenous Indigenous Non-Indigenous Indigenous Indigenous 96.4 3.6 4.1 95.9 Non-Indigenous 0.0 100.0 0.0 100.0 Non- Indigenous Indigenous 100.0 0.0 45.1 54.9 Non-Indigenous 45.5 54.5 0.1 99.9

Parents’ Indigenous status for 5–19-year-olds in 2011 (%*)

* Weighted totals can be found in Appendix Table A.3.
Interpretation: 96.6% of the 5–19-year-olds identified as Indigenous in 2006 and 2011 and who had a mother identifying as Indigenous in 2006 also had a mother identifying as Indigenous in 2011 (Always Identified). The remaining 3.4% of 5–19-year-olds who identified as Indigenous in 2006 and 2011 had a mother identifying as Indigenous in 2006 and as non-Indigenous in 2011.
Note: Figures in bold indicate that a parent’s transition matches the child’s transition.

4 – Transitions and exogamy

28This section continues to assess the Indigenous status of children over the two census time points, but it explores whether transitions are more common in families of mixed (Indigenous and non-Indigenous) parentage. Tables 3 and 4 show children’s Indigenous status in 2006 and 2011 by parents’ Indigenous status in 2006 (Table 3) and in 2011 (Table 4). Newly Identified children are predominately (79.6%) [5] in families with an Indigenous and non-Indigenous parent in 2011. However, the same group in 2006 (who were non-Indigenous in 2006) were overwhelmingly in non-Indigenous-only families. Taken together, these observations suggest that although Newly Identified children are likely to have an Indigenous and a non-Indigenous parent, the Indigenous parent tends also to be Newly Identified across the two censuses. Changes in a child’s Indigenous status should therefore be attributed not only to their status being reported by different parents but also to a parent’s own new identification as Indigenous in 2011. In 2006, Formerly Identified children typically (86.0%) have at least one Indigenous parent in 2006, but in 2011 both parents report overwhelmingly (95.2%) being non-Indigenous. Again, this suggests that children’s status corresponds with that of their parents across time. There are proportionally more Formerly Identified children whose parents were both Indigenous in 2006 (22.6%) than Newly Identified children whose parents were both Indigenous in 2011 (9.3%). That is, while both groups’ Indigeneity aligns with that of at least one parent at a given time point, the Formerly Identified are less likely to be from families of mixed Indigenous status, suggesting that a group of Indigenous children (with two Indigenous parents) in 2006 may be falsely linked to non-Indigenous children (with non-Indigenous parents) in 2011.

Table 3

Children’s Indigenous status in 2006 and 2011 by parents’ Indigenous status in 2006 (%)

Child’s Indigenous status 2006 Child’s Indigenous status 2011 Father’s Indigenous status 2006 Mother’s Indigenous status 2006 Indigenous Non-Indigenous Indigenous Indigenous (Always Identified) Indigenous 37.1 26.8 Non-Indigenous 31.4 4.7 Non-Indigenous (Formerly Identified) Indigenous 22.6 22.9 Non-Indigenous 40.5 13.7 Non-Indigenous Indigenous (Newly Identified) Indigenous 0.0 3.6 Non-Indigenous 3.5 92.1 Non-Indigenous (Never Identified) Indigenous 0.0 0.1 Non-Indigenous 0.1 99.8

Children’s Indigenous status in 2006 and 2011 by parents’ Indigenous status in 2006 (%)

Interpretation: Of the children who were Always Identified as Indigenous, 37.1% had an Indigenous father and mother in 2006.

Table 4

Children’s Indigenous status in 2006 and 2011 by parents’ Indigenous status in 2011 (%)

Child’s Indigenous status 2006 Child’s Indigenous status 2011 Father’s Indigenous status 2011 Mother’s Indigenous status 2011 Indigenous Non-Indigenous Indigenous Indigenous (Always Identified) Indigenous 36.7 24.5 Non-Indigenous 31.8 6.7 Non-Indigenous (Formerly Identified) Indigenous 31.8 6.7 Non-Indigenous 3.2 95.2 Non-Indigenous Indigenous (Newly Identified) Indigenous 9.3 36.9 Non-Indigenous 42.7 11.5 Non-Indigenous (Never Identified) Indigenous 0.0 0.2 Non-Indigenous 0.1 99.8

Children’s Indigenous status in 2006 and 2011 by parents’ Indigenous status in 2011 (%)

Interpretation: Of the children who were Always Identified as Indigenous, 36.7% had an Indigenous father and mother in 2011.

5 – Geographic distribution

29The Australian Standard Geographic Classification categorizes one’s address of usual residence into five groups: major cities/urban, inner regional, outer regional, remote, and very remote. [6] Figure 2 gives the distribution of the four Indigenous groups across these geographic categories for usual residence in 2006. Most of the Indigenous and non-Indigenous population live in major cities. Of those in remote and very remote areas, the majority are Indigenous and are likely to report their Indigeneity on the census consistently. Compared with the Always Identified, those who change status (both Newly and Formerly Identified) are more likely to be living in a city or an inner regional centre. There is little geographic difference between those newly and those formerly identifying as Indigenous.

Figure 2

Distribution across geographic categories of usual residence in 2006 of the four Indigenous classifications

Distribution across geographic categories of usual residence in 2006 of the four Indigenous classifications

IV – Modelling identification change at the individual level

30Along with other components of population change (births, deaths, and geographic mobility), knowing the characteristics that predict identification change is vital for building accurate population models into the future. Taylor (2013) showed that the “forecast accuracy”, or ratio of Indigenous projections to estimates, varied considerably across census years, suggesting there are drivers of population change beyond births and deaths, and which include ethnic mobility. However, there are also likely to be drivers of ethnic mobility that are consistent across years but which vary within the Indigenous population. Here, individual data from the ACLD is informative and fits within an existing literature.

31Caron-Malenfant et al. (2014) used logistic regression to analyse the factors associated with four types of ethnic mobility in Canada: from North American Indian (2001) to non-Aboriginal identity (2006); from Métis (2001) to non-Aboriginal identity (2006); from non-Aboriginal identity (2001) to North American Indian (2006); and from non-Aboriginal identity (2001) to Métis (2006). The first two groups correspond to the Formerly Identified group in this study, while the latter two correspond to Newly Identified. The authors found that 29% of the Métis population and 9% of the North American Indian population in 2006 were newly identified and that identification change contributed the most to population growth. They found that living in an urban area and having a high school diploma were positively associated with the first two identification changes but negatively associated with the latter two.

32Table 5 summarizes a similar set of analyses for the Australian population between 2006 and 2011. Two sets of analysis are summarized in the table. The first is the probability of a person being identified as Indigenous in 2011 who was not identified as Indigenous in 2006 (the Newly Identified). The second is the probability of a person identified as Indigenous in 2006 not being identified as Indigenous in 2011. The first model pertains to the total relevant sample of those aged 5 years and over in 2006. [7] The second model is estimated for those aged 15 to 59 years in 2006 only. This second model allows adjustment for education and labour market characteristics as additional predictors in the model.

Table 5

Relative risk of changing Indigenous status between 2006 and 2011, by Indigenous status in 2006

Explanatory variables Probability of changing from non-Indigenous to Indigenous Probability of changing from Indigenous to non-Indigenous Total People aged 15–59 years in 2006 Total People aged 15–59 years in 2006 Indigenous status in 2006 Not stated vs not Indigenous (Ref.) 10.7737 *** 9.9752 *** n/a n/a Torres Strait Islander or both Aboriginal and Torres Strait Islander vs Aboriginal only (Ref.) n/a n/a 1.8889 *** 1.8627 *** Age in 2006 5–14 4.5817 *** n/a 2.5468 *** n/a 15–24 (Ref.) 25–34 1.4632 *** 1.7710 *** 1.2933 * 1.2573 n.s. 35–59 1.3833 *** 1.2446 ** 1.7937 *** 1.6897 *** 60 plus in 2006 0.8521 n.s. n/a 2.5048 *** n/a Female vs Male (Ref.) 0.9256 n.s. 0.9163 n.s. 0.8233 ** 0.7292 *** Changed usual residence between 2001 and 2006 vs Did not change usual residence between 2001 and 2006 (Ref.) 1.2108 *** 1.0523 n.s. 1.3504 *** 1.1660 n.s. Place of residence in 2006 Lived in a regional area 2.1425 *** 1.8437 *** 0.7021 *** 0.7683 ** Lived in a remote area 5.3092 *** 4.5657 *** 0.1905 *** 0.1832 *** Lived in an urban area (Ref.) Non-Indigenous partner in 2006 (including partner’s status not stated) 0.7045 *** 0.7890 *** 0.9265 n.s. 0.8426 n.s. Indigenous partner in 2006 5.0780 *** 3.3263 *** 0.5528 *** 0.4602 *** No partner in 2006 (Ref.) Not employed vs Employed (Ref.) 1.3002 *** 0.8575 n.s. Completed Year 12 vs Did not complete Year 12 (Ref.) 0.3750 *** 1.0530 n.s. Probability of the base case 0.0013 0.0019 0.0456 0.0598 Number of observations 689,290 447,776 11,946 6,968

Relative risk of changing Indigenous status between 2006 and 2011, by Indigenous status in 2006

n/a: not applicable to the model.
Note: The base-case individual for all models is aged 15 to 24; male; did not change usual residence between 2006 and 2011; lived in a major city in 2006; and did not have a partner in 2006. For the model estimating the probability of changing from non-Indigenous to Indigenous, the base case is defined as someone who stated his or her Indigenous status as non-Indigenous (as opposed to not stated). For the model estimating the probability of changing from Indigenous to non-Indigenous, the base case is defined as someone who identified as Aboriginal only (as opposed to Torres Strait Islander or both Aboriginal and Torres Strait Islander). For the models on the 15–59-year-old population in 2006, the base case is someone who was employed in 2006 and who had not completed Year 12.
Significance levels: *** p < .01, ** p < .05, * p < .10, n.s: not significant.

33Our model development strategy is based on the following hypotheses. First, and most simply, identification change is predictable a priori, and thus all variables are based on 2006 census data. Our second hypothesis is that an individual’s stage within the life course (whether a child, a single adult, a married adult, etc.) is a predictor of identification change, and hence we use age (and sex) as predictors. Our third hypothesis is that social setting matters; we therefore use the Indigenous own detailed Indigenous status, the status of the person’s partner (if partnered), the category of residential remoteness in 2006 (as a proxy for the proportion of the area that is Indigenous), and change in usual residence leading up to the 2006 census (as a proxy for changing social setting). Specific variables are constructed:

Conclusion

48As a new data source containing the largest longitudinal sample of Indigenous Australians, the ACLD provides several new insights. First, the analysis presented here confirms quantitatively, using individual-level data, that Indigenous identification is not necessarily a fixed construct (long-noted in the literature). For some people, Indigenous identification changes over time, which may reflect a change in identity or in the willingness to identify, or simply a change in who completes the census form.

49New identification appears to account for a considerable proportion of the growth in the Indigenous population between 2006 and 2011. Change in identification is not the main factor driving population growth, but it is a factor – there were more new identifications between 2006 and 2011 than there were people ceasing to identify in the period. Assuming there is no net effect of false links on population change, 18% of the increase in Indigenous ERP is attributed to change in identification.

50The group who ceased to identify as Indigenous is difficult to interpret. Children in this group were likely to have both parents identify as Indigenous in 2006 and neither parent identify in 2011. Our analysis suggests this group may contain a relatively high proportion of false links, namely Indigenous persons in 2006 incorrectly linked to non-Indigenous persons in 2011. If this is correct, it would imply that ceasing to identify as Indigenous is less common than observed in this dataset, for the 2006–2011 period, implying that new identifications are a larger driver of population growth than the ACLD suggests.

51Analysis of age groups and the individual-level data show that transitions are most likely to be observed in children. This may be unsurprising as children do not self-enumerate on the census form. However, analysis of their parents’ Indigenous status suggests they come predominately from families comprising one Newly Identified and one Never Identified parent.

52The Newly Identified group appear to possess demographic characteristics that are different from those who consistently identified as Indigenous across the two time points. They were more likely to live in urban areas (and very unlikely to live in remote communities) and had higher socioeconomic status.

53There are two key implications worth highlighting. First, if, as this analysis suggests, Indigenous status is not fixed, repeated cross-sectional analysis of the Indigenous population may yield false conclusions regarding changes over time in outcomes. The fact that people who newly identified as Indigenous in 2011 had higher socioeconomic status than the rest of the Indigenous population would raise the average household income of Indigenous persons in 2011 when compared to the income in 2006. Such changes may be driven predominately by the person who identifies at a given time point, rather than by any improvements in Indigenous well-being. From a policy perspective, governments and researchers need to be much more circumspect when basing decisions and evaluations on data that uses Indigenous population projections, and their conclusions should be based on data that tracks individuals through time. Longitudinal data in which individuals are tracked over time – not just linked – are crucial, given the scale of investment in improving Indigenous outcomes. Second, demonstrating Indigenous identification as a fluid construct, along with the implications for analysis of the Indigenous population, warrants further study into factors that cause people to change how they identify. Such understanding may lead to better analysis of Indigenous populations and perhaps also to better data collection.

54At the time of this study, the ACLD contains only two time points. It is difficult to tell from this dataset what drives change in identification. Given that the greatest rates of change are observed in children and young adults, it follows that changes in identification may be associated with life-course events, including beginning to self-identify, as opposed to a parent or guardian identifying on a child’s behalf. Becoming responsible for one’s own identification relates to a range of possible theories of identity formation and change which require further exploration in the Indigenous Australian context. These include the “looking-glass self” perspective, in which classification by others based on phenotypical appearance is a primary driver of self-identification, and the “presentation of self” model in which physical dress, linguistic, or other cultural cues signal to others and determine one’s Indigeneity in social contexts (Saperstein and Penner, 2014). In addition to such intra- and interpersonal dynamics, one-off factors, such as the 2008 government apology to the Stolen Generations of Indigenous Australians, may also contribute, in which case rates of new identification may not necessarily persist. Whether identification change relates to migration is another area of future research.

55Identification change is not limited to the Australian context. The findings here may well extend to any self-identified Indigenous population to some extent, though the drivers of identification change will be present to different degrees in other populations. While identification change is not a new phenomenon, the ACLD has facilitated the clearest demonstration of it in the Australian Indigenous population.

56The analysis presented here focuses on changing identification as a driver of population change along with a range of new analyses possible in a longitudinal dataset with over 20,000 Indigenous records. However, as with any collection, there are data-quality issues that require careful attention. Here, the Indigenous population had a lower link rate (69% compared with 80% across the whole sample), and the Formerly Identified group appeared to include a higher proportion of false links. Though the presence of false links needs to be considered in interpreting the analysis, it does not preclude making inferences from the dataset. This study has demonstrated a change in identification, and has shown that it contributed to the observed growth in the Indigenous population and that newly identifying Indigenous persons have demographic characteristics that differ from those of the Always Identified population. Another wave of ACLD data (or some other data source) is necessary to determine whether such trends will continue and to help distinguish demographic from other factors (such as one-off events) that contribute to changing identification.