|Page tools: Print Page Print All|
To begin with, the method employed to obtain information in the Australian census is one of self-enumeration in which each household is asked to complete the census form with relatively little assistance from the Census Collector. To make sure that this approach is successful, a series of tests is conducted before each census to gauge public reaction to the form and the questions in it. (See the earlier chapter on Planning the 1996 Census.)
As well as making sure that the right questions are asked, it is essential for the quality of census data that everybody understands the importance of being counted and of giving the right answers in the census. A crucial factor in this respect is the public awareness campaign referred to in the earlier chapter on Collection.
Once the forms are in the Data Processing Centre, quality assurance procedures are implemented to maximise the accurate recording of information collected and to eliminate as far as possible any inconsistencies in coding responses. For this purpose, a sample of forms are recoded by a separate group of coders and the answers compared. Inconsistencies are examined by expert coders and an attempt is made to determine the source of the error. This information along with query resolution information and reports from coders is examined by continuous improvement teams, who have the responsibility of identifying quality problems and ways in which quality can be improved. Coding procedures, indexes, processing systems and training of staff are the key areas where changes can lead to improved data quality during processing.
Despite these efforts, the census, like all statistical collections, is subject to a number of sources of error, and some of the errors defy detection and correction. Testing has indicated the effect of these errors is generally slight, although it could be more significant for analyses of data for small groups or very detailed cross-classifications.
EVALUATING AND OUTCOME
After the census, an evaluation of the quality of census data is carried out to inform users of the data about its quality, and to help plan the next census. Investigation of the effect of partial response, consistency checks between related questions, comparisons with data from other sources and demographic analysis are all carried out for various census topics.
Much of the information gathered about the quality of census data will be distributed in the form of commentary contained in census output products or in specialised data quality evaluation working papers. This information also helps the ABS plan a better census next time around.
SOURCES OF ERROR
Despite efforts to obtain full coverage of people and dwellings, it is inevitable that a small number of people will be missed and some will be counted more than once. In Australia more people are missed from the census than are counted more than once. The net effect when both factors are taken into account is referred to as undercounting.
As well as affecting the total population counts, undercounting can bias other census statistics because the characteristics of missed people are different from those of counted people. In Australia, rates of undercounting vary significantly for different population groups depending on factors such as age, sex and geographic area. A measure of the extent of undercounting is obtained from a sample survey of households undertaken shortly after the census, called the Post Enumeration Survey. Undercounting of people in the 1991 Census was estimated to be 1.8% for Australia as a whole on a place of enumeration basis.
People who are counted in the census do not necessarily answer all the questions which apply to them. While questions of a sensitive nature are generally excluded from the census, all topics have an element of non-response. However, this element can be measured and is generally low. In those instances where a householder does not provide a response to a question, a 'not stated' code is allocated during processing, with the exception of non-response to age, sex, marital status and the statistical local area of usual residence. These variables are imputed, using other information on the census form and specially constructed random tables based on the distribution of the population according to these variables at the previous census. In addition, variables such as birthplace are derived where the appropriate response is clear from other information on the census form.
Computer-editing procedures are used to detect and correct obvious errors made by individuals in completing the form (for example, a six-year-old person with children). However, such procedures cannot detect and correct all householders' errors and some remain in final output.
Errors created during the processing of the census are kept at an acceptable level by means of quality assurance procedures, which involve sample checking during coding operations, and taking corrective action where necessary.
INTRODUCED RANDOM ADJUSTMENTS
Minor adjustments are made to the data to allow the maximum of detailed census data to be released, while at the same time protecting the confidentiality of information about individuals. For this reason, and since possible respondent and processing errors have greatest impact on them, great care should be taken when interpreting data in small cells.