Understanding Census data quality
There are a number of ways in which the quality of Census data can be assessed. Response rates (for both dwellings and people) are measures that are used internationally and are an important part of understanding quality, as is item non-response. In addition, the Post Enumeration Survey provides an independent measure of Census coverage through its estimate of undercount and overcount.
The key indicators for the 2016 Census of Population and Housing support that the Census data is of high quality, with high response rates (95.1%) and low levels of net undercoverage (1.0%).
Dwelling response rate
The dwelling response rate measures the number of private dwellings that returned a completed Census form as a proportion of all private dwellings believed to be occupied on Census night. The dwelling response rate for the 2016 Census was 95.1%. In the 2011 Census this response rate was 96.5%, and in 2006 it was 95.8%. The table below provides a breakdown of dwelling response by state and territory across the last three Censuses.
The decline in the reported dwelling response between the 2011 and 2016 Censuses is partly due to changes in the collection method (see How we collected your information for more details) that resulted in an overestimate of the number of occupied private dwellings. As a result, this has led to a lowering of the overall dwelling response rate. Despite the slight decline, the dwelling response rate is very high, and will produce high quality data.
Dwelling response rate by State and Territory, 2006 - 2016
|New South Wales
|Australian Capital Territory
* Includes Other Territories
Person response rate
The person response rate measures how many people are included on a returned Census form as a proportion of all people (responding and non-responding) believed to be in Australia on Census night. Private dwellings believed to be occupied but did not return a Census form, contribute to the numbers of people who are considered non-responding. Similarly, people believed to be present in non-private dwellings (hotels, hospitals, boarding houses, etc.) but who did not complete a Census form contribute to the numbers of non-responding people.
There are a number of reasons why person non-response occurs in the Census. People may indicate a desire to mail back a Census form or to complete the form online but may forget to do so, some people may refuse to complete a Census form, and some may have been left off a form.
The dwelling response rate (outlined in the previous section) is only calculated for private dwellings, while the person response rate includes all people regardless of whether they are in a private or non-private dwelling,
The person response rate was 94.8% in the 2016 Census. In the 2011 Census it was 96.3% and in the 2006 Census is was 95.8%. Like dwelling responses rates, the over identification of occupied homes have led to a lowering of the reported person response rate in 2016 compared to 2011 and 2006.
Where people are believed to have not responded to the Census, data are imputed for them (see Imputation section in Turning your information into Census data) including selected demographic characteristics such as age, sex, marital status and usual address. This ensures a full and accurate dataset is produced.
Person response rate by State and Territory, 2006 - 2016
|New South Wales
|Australian Capital Territory
* Includes Other Territories
Item non-response rates
The majority of item non-response is attributable to the people who did not respond to the Census at all. These people have records created for them (see Imputation section in the Turning your information into data chapter) and whilst a small number of data items are imputed for them (age, sex, marital status and usual address), the remaining data items are left as not stated.
The second and smaller contributor to item non-response is when people return a Census form but may not answer a particular question(s). This is referred to as item non-response (excl. imputed persons).
Item non-response rates are a measure of how many people did not respond to a particular question as a proportion of the total number of people the question was applicable to. In this instance the response is left as not stated.
This graph shows the contribution of imputed records to item non-response, using Country of Birth (BPLP) as an example.
There is a marked difference between item non-response rates for people who completed their form online versus those who completed the paper form. The Census online form makes a subset of questions mandatory which means respondents cannot submit a blank answer to specific questions. The online form is also quicker to complete, and smart features such as using names to assist with subsequent questions and sequencing the respondent only to necessary questions makes the online form easier to complete. Through these factors, people responding online tend to answer more questions.
The increase in online participation in the 2016 Census has pleasingly reduced the number of questions left blank in completed Census forms.
This table shows item non-response (excluding imputed records) for Country of Birth and illustrates the majority contribution to non-response from paper form submissions.
Item non-response (excl. imputed records) by mode of response for Country of Birth (BPLP)
|New South Wales
|Australian Capital Territory
* Includes Other Territories
Item non-response rates are available for each state and territory, and the whole of Australia (including Other Territories) in Item non-response.
The Post-Enumeration Survey
The Census Post Enumeration Survey (PES) is run shortly after each Census as a way to independently measure Census coverage. The PES determines how many people should have been counted in the Census, how many were missed (undercount) and how many were counted more than once or in error (overcount). While every effort is made to eliminate coverage errors in the Census, some undercount and overcount will inevitably occur. Net undercount for any category of person is the difference between the PES population estimate (i.e. estimate of the number of people who should have been counted in the Census) and the actual Census count.
The PES also provides an estimate of Census imputation error; i.e. the difference between the number of people imputed into non-responding dwellings during Census processing and the number of people who should have been counted in those dwellings. It also provides information on the characteristics of those in the population who have been missed or overcounted, including an indication of those characteristics which may have changed or been misclassified between the Census and PES.
Results from the PES are used for the following:
- Estimates of undercount and overcount provide users with an assessment of the completeness of Census counts and a measure of Census imputation performance. As such, undercount and overcount can be used to evaluate the effectiveness of Census collection procedures and data processing, so improvements can be made for future Censuses.
- Net undercount is a key component in rebasing the Estimated Resident Population (ERP) for 30 June of the Census year. Rebasing is done by adding the net undercount from PES to the new Census population counts, before applying further demographic adjustments in order to recalibrate ERP to the latest Census.
Please refer to Census of Population and Housing - Details of Undercount, 2016 for details.
The PES helps to measure the quality of the Census and ensure the quality of Australia’s population estimates.
The net undercount rate in 2016 was 1.0% (equivalent to 226,407 persons; i.e. the PES population estimate was 226,407 more persons than was counted in the 2016 Census, compared with 1.7% in 2011. The lower net undercount rate in 2016 was driven by a large positive difference in the number of people who were counted on multiple Census forms or counted in error and in the number of persons imputed into non-responding Census dwellings in 2016, compared with 2011. These larger changes in overcount (including over-imputation in the Census) have offset the smaller positive difference in the number of people who were missed from responding Census dwellings.
The 2016 PES estimated a net undercount rate of 17.5% for Aboriginal and Torres Strait Islander peoples (equivalent to 137,750 persons who were missed in the Census). This is only slightly higher than 2011, which estimated a net undercount of 114,188 persons, or a rate of 17.2%.
For Country of birth, the net undercount rate for persons born overseas was lower in 2016 (7.0%) compared with 2011 (8.8%), while persons born in Australia had a higher net undercount rate (8.1% in 2016 compared with 6.6% in 2011). The largest changes in net undercount for the ten highest ranked countries (in terms of population residing in Australia) were for China (from 14.9% in 2011 to 6.2% in 2016), India (from 9.7% in 2011 to 5.0% in 2016) and the Philippines (from 9.1% in 2011 to 5.5% in 2016).
Census Independent Assurance Panel
An independent panel of eminent Australian and international statisticians, academics, and state government representatives was established to independently review and assure the quality of statistical outputs from the 2016 Census.
Overall, the panel found 2016 Census data to be fit for rebasing the Estimated Resident Population and having comparable quality to previous Australian Censuses and International Censuses. The panel concluded that 2016 Census data can be used with confidence.
Their report can be found here.
This quality declaration details how the Census addresses each of the elements of the ABS’ quality framework, covering relevance, timeliness, accuracy, coherence, interpretability and accessibility.
For information on the institutional environment of the ABS, including the legislative obligations of the ABS, financing and governance arrangements, and mechanisms for scrutiny of ABS operations, please see ABS Institutional Environment.
The Australian Census of Population and Housing is the official count of population and dwellings and collects details of age, sex and other characteristics of the population.
The Census aims to measure the number and key characteristics of dwellings and people in Australia on Census Night. All people in Australia on Census Night are in scope, except foreign diplomats and their families. Visitors to Australia are counted regardless of how long they have been in the country or how long they plan to stay. Australian residents not in the country on Census Night are out of scope of the Census.
Topics collected by the Census change from time to time. There must be a demonstrated national need for Census data for policy development, planning and program monitoring. Details on the changing content of Censuses from 1911 to 2016 can be found on the downloads tab of Census of Population and Housing: Nature and Content, Australia, 2016. A copy of the 2016 Census Household Form is included in the Appendix to the Census Dictionary, 2016 (cat. no. 2901.0).
While the range of topics covered in the 2016 Census was the same as for the 2011 Census, there were some changes in the way some questions were either asked or collected. Examples include: changing the question format for religious affiliation by moving 'No religion' to be the first response category; country of birth of mother and father questions were amended to allow respondents to provide details of the country of birth of their mother and father (where it is not Australia); the introduction of some targeted supplementary questions for occupation and industry on the online form; and changes to the collection of data related to dwellings for mail-out areas in line with establishment of the Address Register. Updated classifications were used for the coding of geographical units, occupation, industry, cultural and ethnic groups, language, religion and countries. For more detail see the 2016 Census Dictionary.
The Census and Statistics Act 1905 requires the Australian Statistician to conduct a Census on a regular basis. Since 1961, a Census has been held every 5 years. The 2016 Census was the 17th national Census, and was held on the 9th August 2016.
For the 2016 Census there were three main releases:
- An early release of information in April 2017 (8 months after the Census). This release provided a preview of key characteristics of a 'typical' Australian at the national and state/territory level showing what has changed over time.
- The first comprehensive Census dataset was released in June 2017 and provided Community level Census data data for a wide range of topics, including information on small population groups and for small geographic areas such as Suburbs and Local Government Areas.
- Detailed Census data was released in October 2017 and contained data on employment, qualifications and population mobility, including journey to work and internal migration.
An additional wave of Census-related data will be released from 2018 onwards, adding further value to the main data releases in 2017.
The ABS aims to produce high quality data from the Census. To achieve this, extensive effort is put into Census form design, collection procedures and processing. There are four principal sources of error in Census data which quality management aims to reduce as much as possible; they are respondent error, processing error, partial or non-response and undercount. For more detail, see the 2016 Census Dictionary
The Census is self-enumerated, and respondents sometimes do not return a Census form or fail to answer every applicable question. Persons are imputed into dwellings for which no form was returned, together with some demographic characteristics for these people (age, sex, marital status and usual address). These same demographic characteristics are imputed if not provided by respondents on a returned form. However, the majority of output classifications include a 'Not Stated' category to record the level of non-response for that data item. Data quality statements are produced for each Census data item and include the non-response rate for each variable and a brief outline of any known data quality problems, as well as a comparison with the non-response rate for the 2011 Census. Data quality statements are included in Understanding the data
It is important for Census data to be comparable and compatible with previous Censuses and also with other data produced by the ABS and wider community. The ABS, and the Census, uses Australian standard classifications, where available and appropriate, to provide data comparability across statistical collections. These include, for example, standards for occupation and geographic areas. For more details regarding classifications used in the Census, see About Census Classifications
in the 2016 Census Dictionary
, and the relevant entries for each classifications.
The Census provides a wealth of data about the Australian community through a suite of standard products, and data customised for individual requirements. The 2016 Census Dictionary
is a comprehensive reference guide designed to assist users to determine and specify their data requirements, and to understand the concepts underlying the data. It provides details of classifications used and a glossary of definitions of Census terms.
An extensive range of online products are available on the ABS website. ABS works to ensure all products are as accessible and usable as possible, undertaking testing of colours, images, navigation and language.
If the Census information you require is not available as a standard product or service, then ABS Consultancy Services can help you with customised services to suit your needs. Contact 1300 135 070 from within Australia or +61 2 9268 4909 from overseas for all your Census and other information needs. Alternatively, please email email@example.com
. Your inquiry will be referred to an ABS Consultant who will contact you to analyse your needs, discuss options, and provide you with an obligation free quote.