General
Introduction
The Census of Population and Housing is the most extensive statistical
collection undertaken by the Australian Bureau of Statistics (ABS). It
collects a variety of social and demographic information from all members
of the community.
The results of the census are used by government, industry, academics, many other sectors of the community and private individuals for planning and decision making activities that can affect the lives and welfare of all Australians. Census statistics are used in researching many social problems and as a basis for planning by industry and commerce within the community. Census data also provide an essential basis for the preparation of population estimates at the national, State and local government levels.
Census-taking in Australia
Population counts have been conducted from a very early period in Australia's
history. Initially, these counts were solely head counts celled 'musters',
which were important as a means of matching food and other supplies to the
numbers of people needing them. The first regular census was taken in New
South Wales in 1828 using census-taking methods essentially adapted from
methods used in England at that time.
With Federation in 1901, census-taking became the responsibility of the Commonwealth Government. In 1905 the Census and Statistics Act was passed which gave authority to the Governor-General to appoint a Commonwealth Statistician, one of whose duties was the taking of the census.
The Census and Statistics Act 1905 stipulated that a census was to be taken in 1911 and every tenth year thereafter. The Act also stipulated a number of topics which must be asked in each census: name; age; sex; relationship; marital status; duration of marriage; birthplace; nationality; period of residence; religion; occupation; material of outer walls and number of rooms in the dwellings; and allowed for other topics to be included 'as prescribed'. Since 1911 censuses have been held in 1921, 1933, 1947, 1954, 1961, 1966, 1971, 1976 and 1981.
In 1930 the Act was amended to make the year in which the census was to be conducted more flexible. The Act, which had stated that a census be taken in every tenth year, was amended by the addition of the words 'or at such other times as prescribed'.
Since 1961 a census has been held every five years because of the need to collect, more frequently, data that can only be produced by complete enumeration. In 1977 an amendment was made to the Act to have future censuses carried out on a quinquennial basis from 1981 and at such other times as are prescribed.
Selection of census questions
Planning for the 1981 Census began in mid-1977. In November 1977, the public,
businesses and Government Departments were invited to make submissions on the
topics they wished to see included in or excluded from the census. Response was
excellent, with over 1600 topic requests being received by the ABS. Each
submission was carefully examined for census suitability and public
acceptability. A series of census field trials were held in major cities and
country areas of New South Wales and Victoria to test the more feasible
questions suggested by users.
Preliminary proposals for the 1981 Census were published by the ABS in February 1979 in a paper called Preliminary Views on the Nature and Content of the Census. Australia-wide talks were subsequently held with major users and other interested people which led to further testing of selected census questions. The Australian Statistics Advisory Council was consulted throughout this process. Finally, the Government decided that 35 householder's questions (31 person questions and 4 dwelling-related questions) and 3 collector's questions would be included on the 1981 Census schedule. Some of the topics were specified in the Census and Statistics Act 1905 and the rest were specified in Census Regulations which were tabled in Parliament.
Census collection
Tuesday 30 June 1981 was proclaimed as census night. One week before census
night, 26,666 field staff began distributing census schedules to dwellings
in 26,759 collection districts (CD's).
The Census and Statistics Act 1905 specifies the manner in which the census must be collected. Section 10(1) states that 'For the purpose of taking the census, a form called the Householder's Schedule shall be prepared, and left, in accordance with the regulations, at every dwelling throughout the Commonwealth'. The distribution of the census schedule to each household prior to census day and the collection after census day is undertaken by specially trained census collectors. Collectors are trained to provide help on request to any persons who have difficulty in completing the census schedule.
The Australian census uses a self-enumeration approach to obtain information i.e. each household fills in the details required on the census schedule on its own behalf. This method of collecting information determines the type of questions which can feasibly be included on the census schedule. Topics which involve canvassing opinion, rely heavily on memory, require a large number of questions or an excessive amount of explanation are considered unsuitable for a self-enumeration census.
In the census, all members of the community are counted at the dwelling in which they are staying on census night, regardless of where they usually live. Special provision is made for persons who are not in a private or non-private dwelling, e.g. camping out, at work, travelling on long distance trains or buses and those on boats or planes. The census does not count Australian residents who are overseas but does count people who are visiting Australia. Diplomatic representatives, their families and their dwellings are excluded.
In the two week period following 30 June 1981, census schedules were collected by field staff and forwarded to the census Data Transcription Centre in Melbourne for processing.
Census processing sequences
Answers given to census questions have to be converted to a form in which they
can be read into a computer. In some cases this involves coding information
such as family relationship, labour force status and industry of employment,
before it can be transferred to magnetic tape or disc. In due course, after
undergoing a number of edits (tests) aimed at detecting and adjusting for some
common types of error and some other control
processes, a complete sequence of validated
records is held on a series of magnetic storage tapes/discs known as the
Final Unit Record File. The completion of this process marks the end of input
processing.
As each of the State and Territory Final Unit Record Files is completed, output processing is begun. Output processing consists of sorting, summing, averaging and cross-tabulating data items from the unit record file to produce the statistics for the many uses to which census data are put. Since in these processes of analysis and tabulation some very detailed statistics are produced which may inadvertently allow the release of identifiable information, steps are taken to ensure the confidentiality of information received; these steps include the limitation of geographic detail and the introduction of confidentialising adjustments.
Editing and associated procedures
The aim of editing during census processing is to reduce the number of errors
in the data. The kind of errors that editing procedures can detect are limited
to answers which are inconsistent or invalid. No correction is possible for
errors which do not show up in this way.
In the processing of the 1981 Census there was no correction without reference to source documents for items which failed edits. Imputation of missing data was made in respect of only five items age, sex, marital status, birthplace and occupational status and then only when there was sound basis for such action. These are items frequently used in tabulations.
Two types of edits were applied to 1981 Census processing:
(a) Balancing edits were employed to ensure that the total numbers of persons and dwellings in each CD remained consistent at all stages of processing.
(b) Consistency edits were designed to detect responses which appeared to be in consistent with other responses on the same schedule, or in conflict with census definitions or processing rules.
Apparent inconsistencies in the transcribed census schedule records could result from errors by the respondent in completing the schedule, or by errors in coding or transcribing the information onto magnetic tape or disc. Edits were applied to detect such cases as those in which a person was shown as aged less than 15 years and was also shown as having a marital status other than never married; or when stated age less stated duration of marriage indicated an age of less than 15 years. Although the number of edit failures due to respondent error was small, there were cases when, because of the absence of conclusive information, subsequent adjustment of records was necessarily somewhat arbitrary.
In processing the information from census schedules, all data items are fed into the computer in the form of codes. Edits which test processing rules are applied to ensure that these codes fall into the permitted range. For example, the broken sequence of numbers allocated for occupation codes does not include numbers in the range 092-099; any occupation coding in this range would fail the edit and re-coding would be necessary.
Sources of error in the census
In an operation the size of the census there are many ways for errors to find
their way into the final results. As in other areas of statistics, considerable
effort is directed to devising procedures to ensure the highest possible level
of accuracy consistent with constraints of cost and burden on respondents.
While it is clearly not possible to eliminate all inaccuracies, and some
errors will survive in the final results, it is unlikely that remaining errors
would have any significance in aggregated census data.
Major sources of errors in the census are:
(a) Under-enumeration
The census aims at counting every person and dwelling (excluding diplomatic
personnel and their residences) in Australia on census night. While every
effort is made to minimise undercounting in the census, some inevitably occurs
- for example, the inadvertent omission of very young children or the treatment
by the census collector of an occupied dwelling as unoccupied. Refusal by house
holders to complete the census schedule is not a significant cause of
under-enumeration and accounts for less than 0.5% of households. In about 70%
of these cases the number of occupants was able to be estimated by the
collector from information obtained orally from a member of the household or
other persons and this estimate was included in the census count.
To determine the extent of over or under-enumeration in the 1981 Census, a Post-enumeration Survey (PES) was conducted in the third week following census night, by which time all schedules should have been collected.
The 1981 PES involved interviews with a sample of the population from about 35,000 households across Australia. Specially trained interviewers were used and to ensure independence from the census were not permitted to work as part of the census field staff. Respondents in the PES were asked their name, age, sex, marital status and birthplace for matching with information on the census schedule. They were also asked their place of usual residence, where they spent census night, their address before and after census night and any other address where they may have been included on a census schedule. At each of the addresses given, the personal information was matched to census schedules to establish whether a person was counted, or counted more than once.
Results from the PES indicate that 1.9% of people were missed in 1981 compared with 2.7% in 1976. Under-enumeration rates were lower for all States except South Australia.
Estimated resident population figures for local government areas (LGA's), States and Territories, which are available in separate publications, include an adjustment for under-enumeration. However, statistics available from the census are not adjusted for under-enumeration.
(b) Respondent error
The editing procedures previously described are not able to detect all errors
made by individuals in completing the census schedule so that some errors may
survive in final output. For example, if a respondent states his occupation as
a doctor and he is really a clerk, the census coders will code him to the code
for doctor. However, if his occupation is stated as a doctor but his age is
recorded as 4 years, this combination is defined by census processing rules as
unacceptable and will fail a consistency edit.
(c) Processing error
Every attempt is made through quality control procedures to minimise errors
which may be introduced during the processing of the census. The quality
control system is designed to reduce the incidence of errors in statistical
data and provide information for the efficient administration of processing.
By sampling census information at different stages of the coding operation,
quality control ensures that the amount of introduced error is so small that
for most uses it is insignificant.
(d) Confidentialising adjustments
The ABS has a long and continuing history of protecting the confidentiality of
information which it receives from individuals and businesses. The data from
each census schedule are transferred onto computer files without names or
addresses. After the completion of processing, all census schedules are
destroyed by pulping and no record of names and addresses is retained.
On the 1981 Census schedule, households were assured that 'it would be an offence for any information relating to an identifiable person or household to be released'. This assurance is embodied in the Census and Statistics Act 1905 (as recently amended). It has been necessary, therefore, for the ABS to randomly adjust small non-zero cells in the more detailed tabulations released on microfiche and magnetic tape. These adjustments allow the maximum amount of detailed census data to be published, while avoiding the risk of inadvertently releasing information which could be identified as relating to a particular person or household; the adjustments also allow for a greater output of detailed data than would be possible from the use of other means for protecting the confidentiality of census data.
Cross-classified and small area summary tables on microfiche or magnetic tape have been subjected to random adjustment processes which slightly change some of the data in a tabulation in an unbiased manner. The result is that the data for very small cells are insufficiently exact for information relating to an identifiable person or household to be released but the value of the tables is not impaired. In any case, small numbers in the original data may be affected by such things as respondents omitting to answer particular questions or giving incorrect answers, or by the occasional error in coding or transcribing data in the course of census processing.
Further information on this introduced error is contained in Census 81 - Effects of Introduced Random Error (2156.0).
Intercensal comparability
While many of the characteristics collected in the 1981 Census are similar to
those collected in 1976 and some previous censuses, caution should be
exercised when making intercensal comparisons. Changes in concept, question
wording, respondent attitude, classification, collection methods and money
value can all affect comparability.