3.7 Data items and key population groups

In this section, attention is given to the accuracy of reporting of individual questions (data items). The criterion used to assess the data is whether the 2021 Census data item is as fit-for-purpose (or of similar quality) as the same data item in previous censuses. To do this, the Panel used three approaches:

• comparison of results of the 2021 Census with results for the same data item from previous censuses;
• where available, comparison of the results of the 2021 Census with some other independent data source for the same data item; and
• examination of the 2021 Census results to see if they look broadly reasonable based upon expectations.

There are specific considerations that arise in relation to the assessment of the validity of responses to survey and census questions. Five such considerations are examined in relation to the 2021 Census questions, including:

• the degree of subjectivity of the question;
• the knowledge required to answer the question or the extent to which a person’s characteristics are acknowledged;
• the compulsory or voluntary nature of the question;
• changes in the response boxes to the question; and
• the complexity of what is being measured or the precision of the measurement being made.

The first consideration is that Census questions vary in their degree of subjectivity. For example, the Age data item is objective because most but not all people know their age accurately. At the other end of the spectrum are more subjective questions such as Need for assistance which can be interpreted in different ways or community perceptions can change across time.

A second consideration is the knowledge required to answer a question. The Census form may be separately completed by each individual within a household, or one person may complete the form on behalf of others within the household. Some questions, such as Long-term health conditions, involve information that may not be readily known by others in the household. It is also possible that an individual’s response to a question may change across censuses because they gain knowledge that they did not have at the previous census, such as their ancestry.

A third consideration is that some census questions are voluntary. These items are Religion, and whether the record can be kept in the National Archives. Across censuses, people may change the extent to which they answer voluntary questions, perhaps because of societal changes such as privacy concerns.

The fourth consideration is that census questions use a set of tick boxes for the responses. The order of the tick boxes reflects a logical sequence as is used in labour force questions or the incidence observed at the previous census with boxes being ordered from highest incidence to the lowest. Changes to the question layout other than tick-box ordering may affect the data.

The final consideration is the degree of complexity of measures such as income in a census.

The Panel examined responses to the 2021 Census questions in relation to these five considerations, excluding from the analysis people who did not return a Census form at all. The following discussion covers a range of items which the Panel considers of greatest interest using the five considerations outlined above, noting that not all issues are relevant to each item. The discussion also includes the two new questions, Long-term health conditions and Australian Defence Force service, added to the 2021 Census. Additionally, the ABS separately releases quality statements for all data items.

It should be noted that the data analysis in this section excludes imputed people (people who were not included on a Census form) to allow a more meaningful assessment of data quality across censuses. This is particularly important given the decline in imputation between the 2016 and 2021 Censuses, and the allocation of ‘not stated’ values to imputed people for most Census data items. This means that there may be some small differences between the figures in this report and standard Census products such as QuickStats and Community Profiles, and the latter should be used as the source for most statistical, as opposed to data quality, purposes.

The Census asks whether a person has a need for assistance in self-care, mobility and communication because of a disability, long-term health condition or old age. Item non-response has decreased, from 2.2% in 2016 to 1.8% in 2021, which is an improvement in data quality.

Figure 3.7.1 shows that the reporting of need for assistance has remained consistent over time, with very similar patterns by age group across censuses. The proportion needing assistance has increased slightly across most age groups, continuing the pattern from recent censuses. Excluding the oldest age groups, the largest changes were observed for children aged 5-14 years. The findings in this age group were also consistent with results from the Survey of Disability, Ageing and Carers, which showed an increase between 2012 and 2018 in the proportion of children aged 0-14 years with some level of disability.[1]

[1] Australian Bureau of Statistics (2019). Disability, Ageing and Carers, Australia, Summary of Findings, 2018. Retrieved from https://www.abs.gov.au/statistics/health/disability/disability-ageing-and-carers-australia-summary-findings/latest-release#key-statistics

The Census questions on need for assistance are at the more subjective end of the spectrum. Some people may interpret this question in terms of high-end nursing or medical care while others may consider that an occasional lift to the doctor or to the shops constitutes a need for care. Changes over time may also be due to changes in diagnosis and the availability of assistance, which may explain the increases in need for care among children. In addition, the need for assistance questions in the Census are measuring a complex concept using a small number of questions. Testing of this question indicated that it was broadly comparable with the ‘Profound or severe core activity limitation’ in the more comprehensive ABS Survey of Disability, Ageing and Carers.[1]

The Panel notes the subjectivity and complexity of this item, but considers it suitable for the intended use, due to the consistency in results across censuses, comparability with secondary sources and the decrease in item non-response.

[1] Australian Bureau of Statistics (2018). ABS Sources of Disability Information, 2012 – 2016. Retrieved from https://www.abs.gov.au/ausstats/abs@.nsf/mf/4431.0.55.002

Social marital status reports on partnership arrangements for people who are living together, either in a registered or de-facto relationship. The 2021 Census has extremely similar age and sex profiles to the 2016 Census (see Figures 3.7.2A and 3.7.2B), which indicates that this data item is very reliable and is suitable for use.

No changes were made to marriage related questions on the Census form in 2021. In December 2017, amendments were made to the Marriage Act 1961 which legalised same-sex marriage. This has had the expected result in the 2021 Census, with many people in same sex couples reporting that they are married. This provides reassurance that the data reflects societal change.

Country of birth is provided on Census forms by selecting one of the countries of birth listed with text boxes or by writing the name of another country in the text box. There has been an improvement in the proportion of people reporting their country of birth (see Figure 3.3.4).

Country of birth patterns are consistent across the 2011, 2016 and 2021 Censuses. It is possible to compare Census results with a data set independent of the 2016 Census: the unrebased Estimated Resident Population (see Figure 3.7.3). For this data item, there is broad similarity in the numbers across the two data sources. All countries of birth have slightly lower numbers for the 2021 Census than the 2021 Estimated Resident Population, but this is largely attributable to the smaller total population in the Census count (see Appendix B for further information). Observed differences can also be the result of methodological differences between the two collections. For example, the country of birth for New Zealand citizens, obtained from the Estimated Resident Population, has high levels of imputation due to New Zealand’s introduction of biometric passports in 2005 which replaced country of birth with place of birth. This has some impact on comparability, noting that the ABS is able to allocate country of birth to the majority of New Zealand citizens from historical records or information about town or place of birth.

Country of birth is considered relatively simple to answer and objective except for people whose country of birth has changed its name or its boundaries or has been created since the respondent was born. The question maintained the same response options from 2016 to 2021 Census, except for some minor changes in ordering to reflect the most common responses from the 2016 Census and some changes to the online form to improve user experience.

Country of birth is considered fit-for-purpose based on its consistency with previous censuses, lower levels of item non-response, comparability with Estimated Resident Population, and the objective nature of the question. This conclusion is supported by the Post Enumeration Survey results.

Ancestry is part of the cultural diversity suite within the Census and provides information about ethnic or cultural groups to which people feel they have an affiliation. It is subjective by its very nature and a person’s cultural identity may not be well known amongst unrelated household members. As genealogical research increases, more people are becoming more aware of their ancestry. The question allows provision of up to two ancestries, which may limit people with more complex ethnic backgrounds.

Figure 3.7.4 shows that the proportions over time have been relatively stable for the most common ancestries. Patterns observed in previous censuses have been continued in 2021, such as declining Australian ancestry and increasing Chinese ancestry. Of note is the appearance of Australian Aboriginal in the top ten ancestries for the first time. This is associated with a questionnaire change which is discussed in more detail in Section 3.5.3 Ancestry. The drop in the category ‘Australian’ is affected by the increase in the category ‘Australian Aboriginal’.

Unlike most other items, the level of non-response to this question has not declined since 2016 but has remained at a similar level (see Figure 3.3.5). This maintenance of response levels, along with consistent patterns across previous censuses, indicates that the Ancestry data item is suitable for use after taking into account the increase in the category of ‘Australian Aboriginal’.

The proportion of people using broad language groups at home has remained relatively consistent across the 2011, 2016 and 2021 Censuses, with some gradual change associated with changes in migration patterns (see Table 3.7.1). Unlike most other items, the level of non-response to this question has not declined since 2016 but has remained at a similar level. This question is relatively objective and simple to answer. The data is considered suitable for use.

There were some changes to the 2021 Census form to encourage reporting of Australian Indigenous Languages, however this did not lead to a proportional increase in reporting of these languages, either for the general population or Aboriginal and Torres Strait Islander peoples.

The proportion of people providing a response to the Religious affiliation question has increased from the 2016 to the 2021 Census. Religious affiliation is the only question where the Census form notes that it is optional.

Noteworthy is the continuing increase in people stating they have no religious affiliation. Between the 2011 and 2016 Censuses, the ‘No religion’ tick box was moved from the bottom to the top of the list of common religions on the Census form. At the time, it was not clear whether the rise in ‘No religion’ responses between 2011 and 2016 Censuses was influenced by the change in the order of the tick boxes. However, the proportion of people reporting no religious affiliation increased at a similar level between the 2016 and 2021 Censuses as between 2011 and 2016 Censuses, as shown in Figure 3.7.5. The ‘No religion’ tick box did not change its position for the 2021 Census, which indicates that this is a real change rather than a consequence of questionnaire design and provides reassurance about the quality of data for Religious affiliation.

Income is a complex concept and entails some subjectivity, such as whether to include various sources of income such as superannuation and dividends. Information about income levels might not be commonly shared amongst household members, such as in group households, affecting Census forms where one person completes the form on behalf of all others in the household. It is also regarded as more sensitive information. These factors influence the relatively high item non-response for Income. However, non-response levels were lower in the 2021 Census than 2011 or 2016 (see Figure 3.3.5).

The 2021 Census income distribution is consistent with recent censuses, with the expected move from lower to higher income ranges. In the Census, a single question that groups responses into relatively wide categories is asked. In the ABS Survey of Income and Housing, numerous questions relating to each specific source of income are asked. Given these differences, it is perhaps surprising that the 2019-20 Survey of Income and Housing and the 2021 Census show quite similar income distributions once the not stated responses are removed from the Census data (see Figures 3.7.6 and 3.7.7).

The main differences between the two data sources are for those with no income or a low income, with a very similar pattern observed in 2016. As the Survey of Income and Housing asks about every possible source of income, it draws a response for those with low incomes that the Census does not draw. In addition, the 2021 Census was conducted during the pandemic whereas the 2019-20 Survey of Income and Housing was largely conducted pre-pandemic. Further analysis is needed to understand the interaction between the pandemic and income data, noting that various government benefits and payments were available. The income of people in particular industries such as retail and hospitality may have been more severely impacted by lockdowns. As this issue was foreseen by the ABS, the online Census form included the instruction for people in lockdown: ‘Please reflect your usual income, as it was before the commencement of the current COVID lockdown period’.

Census income data is considered fit-for-purpose and follows the expected income distribution when compared with recent censuses and an independent data source. There is a marked decline in non-response to this question and continues the decline over time.

‘Long-term health conditions’ was a new question for the 2021 Census which asked people whether they have ever been told by a doctor or nurse that they have any selected long-term health conditions. At 4.1%, a relatively high number of people did not answer this question, most likely due to sensitivity issues and a desire not to disclose a condition to others in the household or a possible lack of knowledge of others’ health conditions in the household.

Figure 3.7.8 compares 2021 Census results with the 2017-18 National Heath Survey, for conditions which are contained in both collections. The broad patterns are similar between the two collections. As expected, asthma was a leading health condition among children aged 0-14 years. Men were much more prone to heart disease while arthritis was more prevalent among women. Some health conditions had different levels between the Census and the National Health Survey. Cancer was reported at a higher rate in the Census than reported in the National Health Survey. It is not clear why this is the case and should be subject to further investigation. The most notable difference was for mental health conditions, where the proportion reported in the National Health Survey is substantially higher than that in the Census. Methodological differences between the two collections influence these differences, particularly the level of detail contained in the questionnaires. For example, the collection of mental health in the National Health Survey included prompt cards and questions about a range of mental health conditions, including alcohol and drug problems, mood (affective disorders), anxiety related disorders, problems of psychological development, behavioural, cognitive and emotional problems and other mental and behavioural conditions. In contrast, the Census had a single Census response option of ‘Mental health condition (including depression or anxiety)’ with more subjectivity exercised by respondents.

More investigation is needed into the benefits and limitations of long-term health conditions information collected in the Census. However, more specialised data sources such as the National Health Survey may be better sources for levels of specific conditions and the Census would be more useful for analysis on small geographic areas or cross-classification with other socio-demographic characteristics, taking into account the reporting issues with cancer and mental health conditions.

‘Australian Defence Force service’ was the other new question added to the 2021 Census. It asked whether people were currently or had previously served in the Australian Defence Force Regular and Reserves Services. The level of non-response is considered reasonable (1.9%).

The Census figures for current service in the Regular Service are very close to Department of Defence workforce counts for the whole of Australia (see Figure 3.7.9). Data from the two sources align closely for most states and territories, with relatively small differences in New South Wales, Victoria and the Australian Capital Territory. These differences are potentially explainable by cross-border issues associated with counting of particular bases and people living across a state or territory border from their base. The differences may also have been affected by domestic redeployment for the Defence Force’s Operation COVID-19 ASSIST. The Census count of people currently in the Reserves Service is slightly lower than Department of Defence figures.

The Panel has limited capacity to assess the quality of Census data on previous Australian Defence Force service, as there is neither a definitive independent source nor historical Census data. The Census counted 496,276 people who had previously served in the Regular or Reserves Service.

This consistency with Department of Defence workforce counts and the acceptable levels of item non-response indicate that this Census data item is fit-for-purpose.

The Census form includes a voluntary question about whether each person agrees to their name, address and other information on the form being kept by the National Archives of Australia and released after 99 years. The proportion of people who do not respond to this question has dropped substantially from 10.2% in 2016 to 6.9% in 2021. In addition, agreement to National Archives retention has increased to levels close to those observed in the 2011 Census, following a dip in 2016 (see Figure 3.7.10). Prior to the 2016 Census there was substantial public commentary about privacy of personal information in the Census form, which most likely contributed to a decrease from 2011 in people stating their agreement of their form being kept by the National Archives. The improved response to this question and increased agreement to retention was higher across all states and territories and suggests that the public approached the issue of privacy in the 2021 Census more positively than for the 2016 Census.

Responses to the question on age and date of birth are used to derive the Age item. Age is a fundamental demographic item and date of birth, in particular, is increasingly important in data integration. Item non-response to the Age item is discussed in Section 3.3.5 Item non-response.

Most of the people who responded to this question reported their date of birth, either with or without their age. The 2021 levels for date of birth reporting are higher than those observed in the 2011 and 2016 Censuses (see Figure 3.7.11). The large increase may be explained by reduced privacy concerns and questionnaire changes. For the first time, the 2021 Census asked people to provide their age and date of birth (previously just one was requested), with a note that age should be given if date of birth is not known.

The Panel considers Age data fit-for-purpose, based on increased response to this question, particularly for date of birth. See also Section 3.4 Population counts and age - sex distributions.

At the national and state and/or territory levels, changes observed in the data items reviewed by the Panel align with expectations. Some data items involve more subjectivity or complexity but are suitable for their intended use. Greater levels of response to most questions may indicate a greater willingness to disclose information on the 2021 Census form and have resulted in improved data quality. The data items assessed in this section generally have results which are consistent with previous censuses or independent data sources, or have observations consistent with expected societal change or pandemic impacts. The ABS should evaluate the two new topics.