2940.0 - Census of Population and Housing - Undercount, 2006  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 05/06/2007   
   Page tools: Print Print Page Print all pages in this productPrint All

EXPLANATORY NOTES


SCOPE OF THE 2006 CENSUS

1 The 2006 Census of Population and Housing was held on 8 August 2006. The objective of the Census was to measure the number and key characteristics of people in Australia on Census night, and the dwellings in which they live.


2 The aim of the Census was to count every person who spent Census night in Australia. This includes Australian residents in Antarctica and people in the territories of Jervis Bay, Cocos (Keeling) Islands and Christmas Island. The other Australian External Territories, Norfolk Island and minor islands such as Heard and McDonald Islands, are outside the scope of the Australian Census. The only people who spend Census night in Australia but are excluded from the Census are foreign diplomats and their families.


3 Visitors to Australia are counted regardless of how long they have been in the country or how long they plan to stay. Australian residents out of the country on Census night are out of scope of the Census. People outside Australia who are not required to undertake migration formalities, such as those on oil and gas rigs off the Australian coast, are included. People were counted where they stayed on Census night. This means that the Census was conducted on an 'actual location' or 'place of enumeration' basis.


4 All private dwellings, except diplomatic dwellings, are included in the Census, whether occupied or unoccupied. Caravans in caravan parks, manufactured homes in manufactured home estates and self-care units in accommodation for the retired or the aged are counted only if occupied. Occupied non-private dwellings, such as hospitals, prisons, hotels, etc., are also included.


5 Details about the 2006 Census content, collection operations, confidentiality and privacy protection, processing and evaluation activities are contained in 2006 Census Nature and Content (cat no. 2008.0).



SCOPE AND COVERAGE OF THE 2006 PES

6 The scope of the Census is every person present in Australia on Census night (with the exception of foreign diplomats and their families). Ideally the PES would sample from all people who were or should have been counted in the Census - thus this set of people constitute the theoretical scope of the PES. For practical reasons there are a number of areas, dwellings and people excluded or not able to be covered by the PES. Of the people present in Australia at the time of the PES, the following are not included:

  • people in non-private dwellings such as hotels, motels, hospitals and other institutions
  • homeless people (as the sample selected in the PES is based on the selection of dwellings)
  • foreign diplomats and their families
  • overseas visitors who were not in Australia on 8 August 2006 (Census night)
  • babies born after 8 August 2006
  • people in Cocos (Keeling) Islands, Christmas Island, Australian Antarctic Territory, and Jervis Bay Territory.

7 The PES also does not obtain information about people who died between Census and the PES. However, it does obtain information about Australian residents who are overseas during the PES enumeration period and who departed some time in August, provided that they usually live with people remaining in Australia (in private dwellings).


8 The 2006 PES included remote areas and discrete Indigenous communities for the first time. Previous PESs have excluded these areas from the coverage of the survey due to operational reasons - mainly the additional cost and the need to use the same local contacts as Census, which was considered likely to compromise the independence of the PES. Inclusion of these communities in 2006 ensures the geographic coverage of the PES is more complete than it has been in the past. In practice, the PES is used to produce estimates for the full Census scope, even though its actual coverage is somewhat less.


9 The PES interview process determines whether each person in the sample should have been counted in the Census, and a few of the various categories in which they should have been counted (such as age, sex, Indigenous status, country of birth, region of usual residence, etc). The match and search process involves comparing dwellings and people enumerated in the PES with dwellings and people counted in the Census. The objective is to determine how many times each person in the PES sample was actually counted in the Census. PES output processing and estimation combines and weights results from the match and search process to produce an estimate of the number of people who should have been counted in the Census.



PES ESTIMATION

10 Following the 2001 PES, a review of the PES estimation method was commissioned to develop an estimator for the PES that adjusts adequately for non-response and non-coverage in PES, and for miscounting in the Census. The estimator used in the 2001 PES did not fully account for people missed in both Census and PES.


11 One outcome of the review was the development of a new estimator, Prediction Regression (PREG), for use in the person weighting stage of the 2006 PES. Unlike the estimator used in 2001, the weight adjustment applied by PREG to each PES person does not depend on their Census response. The PREG estimator also allows for differences in reporting of a person's characteristics (e.g. age, Indigenous status) between PES and Census and enables a correction for errors in Census imputation.


12 Another outcome of the review was an improved approach to adjusting for dwellings found in the Census but not responding in the PES. This and other changes described in this publication had a larger impact on the final estimates than the change in estimator.


13 The 2006 PES introduced a number of methodological improvements that allow more aspects of the Census to be appropriately measured. A side-effect of these changes is that the sampling error on the overall population estimates has increased, with previously unmeasured potential errors now being measured by the survey and included in adjustments.


14 Further details of PES estimation are in Appendix 3. Technical details of the PREG estimator were reported in Research Paper: An estimating equation approach to Census coverage adjustment (cat. no. 1351.0.55.019) released on 7 May 2007. Further descriptions of PES estimation can also be found in Information Paper: Measuring Net Undercount in the 2006 Population Census, Australia (cat. no. 2940.0.55.001) released on 7 May 2007.



CENSUS COUNTS

Usual residence

15 The Australian Census counts people where they actually were on Census night, rather than where they usually live. There is, however, a need for data based on place of usual residence, and Census counts are available on this basis.


16 For usual residents of Australia, 'place of usual residence' for the 2006 Census is defined as the address at which a person has lived or intends to live for six months or more in 2006. While for most people their usual residence was the same as their actual location on Census night, some people spent Census night at a place other than where they usually lived. Thus, their 'place of enumeration' and their 'place of usual residence' were different.


17 People visiting Australia on Census night are included in the Census counts on a place of enumeration basis but not those on a place of usual residence basis.


18 Usual residents of Australia who are temporarily overseas on Census night are not included in Census counts on either a place of usual residence or place of enumeration basis. However, counts of these people are accounted for in the estimated resident population of Australia (ERP). For information on the calculation of ERP, see the ABS publication Australian Demographic Statistics, December quarter 2006 (cat. no. 3101.0), released on 5 June 2007.


19 Estimates presented in this publication are on a place of usual residence basis.


Census late return and imputed dwellings

20 For some people who have not returned a Census form, contact from the ABS following selection in the PES acts as a reminder and possible motivator to return a completed Census form. These late returns, if not identified, would result in the PES sample having a higher proportion of Census response than in the overall population. To protect against this, all Census forms received after the start of PES field work are deemed 'late'. For the purpose of PES estimation, the dwellings from which these forms are received are treated as though they had not been contacted in the Census, and are classified to the 'non-contact sector' of the Census.


21 The non-contact sector also contains dwellings which were non-responding in the Census - that is, dwellings where the Census never obtained a return, and which could not be established as having been unoccupied on Census night. These non-response dwellings are given imputed values (using 'hot-deck' imputation) during Census processing, based in many cases on information provided by the Census collector about the dwelling and its residents. Inevitably, the imputed values, at the dwelling and aggregate level, differ from the true, but unknown, values. The imputed records constitute the majority of the Census non-contact sector records; late returns (as defined here) are only a small component of the overall Census non-contact sector. Given that late returns prompted by PES would otherwise have been classed as non-response in the Census, the PES sample is representative of the whole non-contact sector, even though it cannot split late returns from non-responses in a manner comparable to the Census.


Resolution of Census not-stated values

22 For Census purposes, age, sex, marital status and state of usual residence are imputed (statistical process for predicting values where no response was provided) during Census processing where these items have been left blank, including where a whole person record has been imputed. Missing values for any other items remain 'not-stated' in the final version of Census counts.


23 The PES uses Census data items to form benchmark categories for weighting and estimation purposes. Two such data items are Indigenous status and country of birth. In cases where these items have been left blank in the Census, a value was imputed during PES processing so that these items could be used for 2006 PES benchmarks.


24 The imputation method used involved imputing both variables together. This enforced the assumption that anyone who is imputed as Indigenous is born in Australia.


25 For benchmarking purposes, 12 categories of persons were defined:

  • Indigenous, Australian-born
  • non-Indigenous, Australian-born
  • non-Indigenous, born in other country - 1 to 9 ranked (in terms of population) countries of birth from the 2006 Census
  • non-Indigenous, born elsewhere.

26 The full range of 10 non-Australian country of birth classes were used in benchmarking at the Australia level; for most regions in Australia they were collapsed to 2 classes.


27 These are non-overlapping categories, but persons with not-stated values in the Census may be imputed as partially in two or more of the categories (with the sum of all categories adding to 1 for each person).


28 Imputation was performed separately for each person within non-overlapping imputation classes, whereby data are imputed by the proportion of stated values of respondents in the same imputation class. In order of importance, these classes were:

  • Statistical Local Area - geographical areas, in almost all cases, identical with whole legal local government areas, comprising cities, district councils, community government councils, municipalities, shires, rural cities, and towns.
  • Census form type
  • Age Group - 5 year age groups to 75+
  • Sex.

Differences in classification

29 Occasionally the answers obtained for a person in the PES interview were not consistent with the answers obtained for the equivalent questions in the Census. There are a number of reasons a response may differ, including:
  • a person may have difficulty answering a question for themselves or another household member, either in the Census or the PES
  • a person may interpret the question differently in the Census, where forms are self-completed for the majority of Australia, than in the PES, which is administered by an interviewer
  • different people may provide the Census and PES answers
  • the correct answer could change between the Census and PES. Changes in age can be taken into account using the actual date of birth, but other changes, for example if the person married or divorced, may not be identified.
  • the Census may contain a 'not-stated', or imputed, response while the PES will have a valid response.

30 The PREG estimator adjusts the weights of responding persons according to their PES-reported categories. This ensures that PES persons not responding in the Census get the same weight adjustments as similar persons who did respond. The Census categories are used in producing sample-based estimates of numbers of persons counted in the Census. The weighting ensures that these estimates match the actual Census counts for all benchmark categories.



CORRECTION FOR CENSUS IMPUTATION ERROR IN THE NON-CONTACT SECTOR

31 The standard error (SE) on the PES estimate of the population in the non-contact sector has been calculated, and used in comparing the accuracy of the PES estimate with that of the unadjusted Census count for this sector. Analysis showed that the overall population estimates are considerably more accurate if the PES estimates are used for this sector rather than using the Census counts (which in this sector were mostly imputed).


32 The decision to use the PES to estimate the population in the non-contact sector allows the sampling error in this sector to be measured and included in the published standard errors. This is the major reason that SEs for the 2006 PES exceed those of previous PESs. The increased SE represents uncertainty in estimating the contribution of the non-contact sector. This is a much better outcome than the alternative of including an unmeasured and potentially major bias arising from unadjusted inaccuracies in the Census imputation process. The ability to measure this sector is one of the key improvements made in the 2006 PES.


33 Net undercount estimates presented in this publication incorporate the PES estimates for the population in the non-contact sector.



INDEPENDENCE FROM THE CENSUS

34 The purpose of the PES is to provide an independent check on Census coverage. There are two aspects to this independence: operational independence and population independence. Operational independence requires that Census operations do not influence the PES in any way, and vice versa. ABS controls this very closely as described below. Population independence means that there should be no subgroups of the population where being missed in the Census indicates that a person or dwelling is more likely to be missed by the PES also. This is harder to achieve, but the PES estimation process can adjust for this to some extent by subdividing the population into smaller groups where the assumption of population independence is more likely to be true.


35 Steps were taken to maintain the operational independence of the 2006 PES from the Census at every stage of the survey, including enumeration, processing and administration. These steps included:

  • selecting the PES sample from an independent sample frame
  • using separate office staff in the PES and Census where possible
  • ensuring the PES interviewers were not employed as Census field staff in the same area, and vice versa
  • maintaining the confidentiality of the PES sample so that Census field and office staff were not aware which areas were included in the PES.

36 For some people who have not returned a Census form, contact from the ABS following selection in the PES acts as a reminder and possible motivator to return a completed Census form. These late returns, if not identified, would result in the PES sample having a higher proportion of Census response than in the overall population. To protect against this, all Census forms received after the start of PES field work are deemed 'late' and treated differently in PES estimation.


37 The PES questions are asked of householders face-to-face by experienced, highly trained interviewers, whereas most Census forms are self-completed. The PES is also a much smaller scale operation (and hence easier to control) than the Census. These features enable the PES to deliver an accurate estimate of the percentage of people and dwellings missed by the Census.


38 The Census can also be used to form an estimate of the percentage of dwellings and people missed by the PES. The PES excludes non-private dwellings (hotels, motels, hospitals) for operational reasons. The PES is also conducted several weeks after the Census, so a respondent's recollection of their location on Census night may not be entirely accurate. Census has special procedures for enumerating homeless people, while the PES is essentially a survey of dwellings and the people who reside in them. Thus the Census may include some dwellings and people that the PES misses. PES estimation implicitly accounts for the dwellings and people missed in the PES but counted in the Census.


39 On the assumption that the Census and the PES are independent, the estimate of the percentage missed by the PES but found by the Census, and the percentage missed by the Census but found by the PES, can be used to construct estimates of the percentage missed by both PES and Census.


40 Despite efforts to maintain independence, the likelihood of a person being missed in the PES may be related to whether they were missed in the Census. This may result in a 'correlation bias' in the PES estimates. To minimise this bias, PES estimation takes account of the fact that different groups have a different likelihood of being missed.



RELIABILITY OF THE UNDERCOUNT ESTIMATES

41 As the estimates of undercount are based on data from a sample survey, they are subject to sampling error. Some of the estimates presented in this publication have high SEs, and these estimates should be used with caution. For more information about SEs see the Technical Note.


42 The estimates of undercount are also subject to non-sampling errors which occur in all collections, censuses and surveys. Examples of this kind of error include imperfections in reporting by respondents, errors made in collection of data, and errors made in processing the data. Every effort is made in the Census and PES to reduce non-sampling error to a minimum by careful design of forms, training and supervision of collectors and interviewers, and by using efficient operating procedures. Types of non-sampling error arising from the way the PES is conducted and the way estimates are derived from the survey are discussed below.


43 A potential weakness in the PES method is its dependence on matching as a means of deciding whether or not a given person or dwelling has been counted in the Census. The difficulties associated with the matching process mean that there is a risk of failing to match people who are actually included in the Census. The effect of not matching when there should have been a match would be to overstate net undercount in the Census.


44 While the Census and PES are conducted independently of each other, they are very similar in many respects. Thus, some weaknesses in the Census may also be shared by the PES leading to an understatement of net undercount. For example, dwellings missed by a Census collector are often difficult to find and so could possibly be missed by a PES interviewer as well. Also, people who avoid being included in the Census may also avoid being included in the PES. The use of benchmarks in estimation is designed to control the effect of this 'correlation bias'.


Census Data Quality

45 Census data are subject to a number of inaccuracies resulting from errors by respondents or mistakes in collection or processing. Whilst many of these are corrected by careful processing procedures, some still remain. The effect of the remaining errors is generally slight, although it may be more important for smaller groups in the population. Using the characteristics reported in PES is intended to correct for such errors.


46 The main kinds of errors occurring in the Census are:

  • Partial non-response: In some cases where an answer is not provided to a question an answer is imputed (often from other information on the form). In other cases a 'not stated' code is allocated.
  • Processing error: While such errors can occur in any processing system, quality management is used continuously to improve the quality of processed data, and to identify and correct data of unacceptable quality.
  • Random adjustment: Table cells containing small values are randomly adjusted or suppressed to avoid releasing information about particular individuals, families, or households. The effects of these changes is statistically insignificant.
  • Respondent error: Because processing procedures cannot detect or repair all errors made by people in completing their forms, some remain in final data.
  • Undercount: Although the Census aims to count each person, there are some people who are missed and others are counted more than once.

47 For further information on sources of error in the Census, refer to the appropriate entries in the 2006 Census Dictionary (cat.no. 2901.0) released on 26 May 2006.