Australian Bureau of Statistics
2963.0 - Census Working Paper 93/4 - Comparison of Census and PES Responses, 1991
Latest ISSUE Released at 11:30 AM (CANBERRA TIME) 06/01/1994
|Page tools: Print Page Print All RSS Search this Product|
Census Working Paper 93/4
1991 PES: COMPARISON OF CENSUS AND PES RESPONSES
Population Census Evaluation
Aboriginal/Torres Strait Islander Origin
Census Data Quality
Census counts are inevitably subject to inaccuracies resulting from errors or omissions by respondents, either unintentional or deliberate. Information
collected as a part of the Post Enumeration Survey (PES) can be compared with information collected in the Census and the consistency of responses
analysed for implications for data quality.
Post Enumeration Survey
The PES is conducted shortly after the collection of Census forms has been completed. A sample of private dwellings across Australia is selected.
At all selected dwellings, interviewers ask Any Responsible Adult (ARA) a series of questions for each person in the household concerning personal
characteristics and location on Census night. If the ARA is unable to provide the information required for each person, another household member may
be asked, or a personal interview may be arranged. There are five questions common to both the Census and the PES: sex, age, marital status,
birthplace and Aboriginal/Torres Strait Islander origin.
There are a number of differences between the two collections which may have affected respondents. The PES is conducted several weeks after
Census night and some differences are inevitable, for example, someone may have been married in that period. In the Census, people completed their
own Census form and thus may have made errors in interpreting the question or may have skipped questions which requested sensitive information. In
the PES, a trained interviewer directly asked householders questions and could prompt them for the relevant information, thus responses may differ and
omissions were less likely.
While PES data is likely to be of a higher quality (because it is collected by a trained interviewer), it cannot be regarded as necessarily 'correct' by
objective standards. Interview responses are still subject to the inaccuracies of recall and insufficient knowledge about other household members,
as well as to the effects of interviewer-respondent interaction. For example, people may feel uncomfortable revealing personal information to a
stranger. While the comparison of results will not show how many people made errors in completing the Census form, it may still reveal information
about the quality of the data obtained in the Census.
The population used in this analysis consists of people in the PES sample who were enumerated in the Census and for whom the extraction of
Census responses from the IFURF was complete.The analysis is based on unweighted data of 86,226 records. Sampling error should be taken into
account when comparing the figures. The data was analysed in a number of ways. Firstly, for people who responded in both the Census and PES,
responses were compared. Secondly, non-response in the Census was examined. Thirdly, where information is available on responses imputed in
the Census (see 1.4), the imputed values can be compared with the responses given in the PES.
Non-response and Imputation
The procedure for imputation in the Census for non-response differs between questions. During pre-capture, a value was derived for sex if it was not
stated. No record is available of how many people did not respond to this question, thus analysis of imputation is not possible. During preliminary
and main processing, responses were imputed where missing for age and marital status and PES information is available on the people for whom
responses were imputed. Some imputation was performed for missing responses for Aboriginal/Torres Strait Islander origin or birthplace but no record
is available on the number of responses involved. The non-response rates stated in this report for these questions refer to those persons for whom no
response was provided and for whom no response was imputed.
In the PES, the interviewer would sometimes estimate the age of the respondent and in the following analysis such cases have been treated as not
responding in the PES.
Table 1 below contains some summary information on the comparison of Census and PES responses for the five questions common to both collections,
and on the non-response rates for these questions in the Census and in the PES.
TABLE 1: Percentage of persons whose Census and PES responses agree and non-response rates in the Census and PES
Discrepancy between Census and PES responses was greatest for marital status and age and least for the questions on sex and Aboriginal/Torres Strait
Islander origin. Most of the differences between Census and PES age were small, however, with 98.15% of persons having an age in the PES which was
within one year of the Census response.
As expected, non-response generally was much lower in the PES than in the Census, with the exception of age. This could be due to the number of people
for whom the interviewer estimated age who have been treated as not responding.
There is very little inconsistency among responses to the question on sex, and the few cases observed are probably due to the imputation performed at
pre-capture. This question will not be examined in more detail in this report.
As noted in the introduction, there was a surprisingly large level of disagreement between age given in the Census and age given in the PES. These
inaccuracies in age data will be explored in detail in this section.
The analysis is complicated by the fact that two questions on age were included in the PES: the first question requests the age last birthday and the
second requests the date of birth. A comparison of responses to these two questions is included as well as comparisons between responses in the
Census and PES. Persons who did not respond in the Census, and so for whom age was imputed, are considered separately in the analysis.
Age in the PES
Over 3.5% of respondents in the PES gave an age and date of birth which did not agree, although most differences were of only one year. Tables 2 and 3
below shows the difference between responses and the non-responses rates to each question. AGE yrs is the age last birthday from the first question and
AGE dob is the age calculated from the date of birth from the second question.
TABLE 2: Response status for AGE yrs and AGE dob in PES
The non-response rate was slightly higher for the date of birth question, indicating that this question may have been slightly more difficult for interviewees to answer.
TABLE 3: Comparison of AGE yrs and AGE dob in PES
* People whose date of birth was between Census night and the final day that the PES was in the field. Depending on the date of birth, the answers of many
of these people could legitimately differ by one year.
Nearly all responses (96.4%) had age and date of birth within one year of each other. Of those respondents whose age from the date of birth was different to
that given, 1.9% gave an age lower than that indicated by their date of birth, while 1.7% gave an age that was higher than that indicated by their date of birth.
For differences greater than 1 year, the difference tended to cluster around 10 and -10.
Comparison of Census and PES responses for age
The 'age last birthday' question included in the PES was similar to the question in the Census, although they were not identical. In the PES the interviewer
recorded the answer while in the Census the respondent had to mark an answer box in a grid.
Table 4 shows the level of agreement between Census and PES responses to both age last birthday AGE yrs and age calculated from date of birth AGE dob
TABLE 4: Consistency of Census and PES responses for age
* People whose date of birth was between Census night and the final day that the PES was in the field are excluded. Depending on the date of birth, the
answers of many of these people could legitimately differ by one year.
Even excluding people whose birthday was between the Census and PES, 8.2% of the population had answers to the two 'age in years' questions which
differed by one year, and a similar level was found comparing the Census age and that from the PES question on date of birth. This high incidence of small
errors may partly be due to the fact that, in both the Census and the PES, one person may respond for the entire household. A respondent may be unsure
of another person's exact age or date of birth, and answers may differ in the two collections, especially if different people complete the Census and PES.
Table 5 compares the Census and PES responses by 5 year age groups. Persons born between the Census and the end of the PES are not included.
The lowest level of agreement (between 96% and 97%) was found for people aged between 40 and 74 while the highest level of agreement was found in the
age group 0 to 4 years, at 99.5%. The other age groups (75 years and over and 5 to 19 years) had levels of agreement between 98.4% and 98.9%.
For people aged between 20 and 70 years, the errors tended to be within the same 10 year age range. That is, for example, a person who gave an age
between 40 and 44 in the Census and whose age in the PES differed was more likely to be in the 45-49 age group in the Census than 35-39. This difference
was often in the range of 1% of respondents.
From an analysis of persons whose age from date of birth data differed from the age given in the Census by more than 50 years, it appeared that some
people gave their year of birth in the Census rather than their age in years. Table 6 includes the details of some of these cases.
TABLE 6: Examples of persons who may have given their year of birth in the Census
A search of the PES file revealed about 51 people (12% of those who had differences of two years or greater) who may have made this error.
Analysis of PES responses for age for cases of non-response in the Census
Non-response to age in the Census for PES respondents was relatively low at 0.51%. The responses recorded in the PES for people for whom no response
was given in the Census are contained in Table 7 below.
TABLE 7: PES responses AGE yrs for all PES respondents and for cases of non-response in the Census
The rate of non-response was highest for those identified as being 15 to 19 years old in the PES, although the reasons for this are not clear. Non-response
in the Census was also high amongst those with their age in the PES as being over 75 years. This could indicate that elderly people had difficulty in
completing the question on age in the Census.
Comparison of PES responses to age and values imputed in the Census
Table 8 contains data for all persons in the PES for whom age was imputed in the Census. The older age groups are combined as the sample sizes are much
smaller and the major point of interest is among the younger age groups.
TABLE 8: PES responses for AGE yrs persons with age imputed in the Census
There is a much lower correspondence between age in Census and age in PES here, as would be expected. Most cells in the table contain very few records
with the exception of those with imputed age in the Census of between 10 and 14 and given age in the PES of between 15 and 19. There were 60 people in
this category, indicating that the imputation system may have given many people ages of younger than 15 years when in fact they were 15 years or older. For
all PES age groups in the table, including the combined 30 years and older group, about 40% had a lower age imputed and only 11% had a higher imputed age.
If Marital Status was not stated in the Census then it was imputed according to dwelling type, income, student status, educational institution attended,
relationship and age. This analysis will compare both the stated and the imputed responses in the Census with the responses in the PES.
Only people identified as being aged 15 years or older in the PES will be considered here. There was very little inconsistency in the remainder as nearly all
children aged under 15 years had a response of 'Never Married'. If a response was not stated in the Census then a response of 'Never Married' was imputed.
It appears that no response was given for many children for whom the question would have appeared irrelevant. The non-response rate for children under 15
was 11.5%, compared to 3.7% for people aged 15 or older.
Comparison of Census and PES responses for marital status
Table 9 shows the extent of consistency between the PES and the Census responses for marital status.
TABLE 9: Consistency of Census and PES responses for marital status
Marital Status has a slightly lower rate of agreement than most other variables. This may be the result of responses varying between the self-enumerated
and interviewer based collections because of the sensitivity of the question for some people.
Table 10 provides more information on the comparison of Census and PES responses, comparing the responses given in the Census and the PES.
The response categories which are subject to the greatest change between the Census and the PES are 'Separated' and 'Divorced'. Of those persons with
marital status of 'Separated' in the Census, almost 10% are shown as married in the PES. Persons who are shown as divorced in the Census and had a
different response in the PES were not concentrated in a particular group, with between 4.3% and 6.2% responding in each of the categories 'Never Married',
'Married', and 'Separated'. While some minor change could be expected between the Census and the PES as persons became married, widowed or divorced,
such changes would not account for the variations shown.
A number of possible reasons for the inconsistencies can be postulated but any conclusions about actual causes are bound to be speculative. Two factors
that are likely to be influential are respondent confusion in the Census (eg. uncertainty between legal marital status and status in a current relationship) and
greater sensitivity towards disclosing legal marital status under some circumstances to interviewers.
TABLE 10: Responses in the Census and PES for marital status
Analysis of PES responses for marital status for cases of non-response in the Census
For PES respondents, marital status had a relatively high non-response rate in the Census, at 3.72%. Table 11 below shows the responses given in the
PES by those people who did not give a response in the Census.
TABLE 11: PES responses for cases of non-response to marital status in the Census
People who were identified as being married in the PES were slightly more likely to have had a response provided in the Census than other people.
Comparison of PES responses to marital status and values imputed in the Census
Marital status was imputed for people who did not respond in the Census. Overall, two thirds of the imputations performed were borne out by the PES.
Table 12 below compares the imputed values with the responses given in the PES.
TABLE 12: PES responses for marital status; persons with marital status imputed in the Census
The imputation algorithm appears to have been able to correctly classify married people, with over 90% of those identified as married in the PES having had a
marital status of married imputed in the Census. The algorithm also appeared to work well for those identified as never married in the PES, with almost 60%
having had that response imputed in the Census. The level of agreement between the imputed Census response and the PES response was much lower for
the other categories of separated, divorced and widowed, which is not surprising given the relatively small number of people in these categories in the population.
The first stage of this analysis compares the responses of those for whom birthplace was stated in the Census with the response in the PES. The second
stage of the analysis considers the PES responses of persons with their birthplace not stated in the Census. The sample size for many countries was very
small and this should be taken into account in drawing conclusions from the following analysis.
The comparability of responses could be affected by the different lists of countries in the PES and the Census. In the Census, there were separate categories
for England and Scotland, while in the PES one category of UK/Ireland was provided. Yugoslavia, Germany and Vietnam were listed in the PES but not in the
Comparison of Census and PES responses for birthplace
The overall percentage of persons whose Census and PES responses agreed was 98.8%. Table 13 compares the responses in the Census and the PES
There was a very high correspondence among those who gave Australia as their birthplace in the Census and the PES, with 99.6% of persons who said they
were born in Australia in the Census giving Australia as their birthplace in the PES.
Between 94% and 99% of persons giving UK/Ireland, Italy, Greece, Netherlands, Germany, New Zealand, Vietnam, Other Europe or Other Asia as their
birthplace in the Census gave the same result in the PES. For many of these countries a common change is to give Australia as birthplace. Other common
responses tended to be in the 'Other Europe' (for European countries), 'Other Asia' (for Vietnam) and 'Other' categories. Of those whose birthplace was coded
as 'Other' in the Census, only 87.6% were coded as 'Other' in the PES. This could be due to slightly different names being used for birthplaces.
The highest rate of change among the countries was observed for those who gave Yugoslavia as their country of birth in the Census, where 9.6% of persons
gave a different country of birth in the PES. This could partly be due to the political instability in that country. A response of 'Macedonia' to a birthplace
question would have been coded to Yugoslavia, according to ASCCSS codes. However there has been contention over the use of the term 'Macedonia' as
some people born in Greece also identify as Macedonian. Over 5% of people who were identified as being born in Yugoslavia in the Census had a birthplace
of Greece in the PES and it appears likely that this is the result of the situation mentioned above.
Analysis of PES responses for birthplace for cases of non-response in the Census
Birthplace was not stated for 2.0% of persons in the Census. Table 14 shows the birthplaces stated in the PES for these persons.
TABLE 14: PES responses for cases of non-response to birthplace in the Census
Those people who didn't respond in the Census were more likely to indicate in the PES that they were born in countries other than those (Australia, England,
Scotland, Italy, Greece, New Zealand, The Netherlands) listed on the Census form. Thus it appears that there was a 'list effect' in the Census and people were
more likely to respond if they only had to mark a box rather than write in a response.
Aboriginal/ Torres Strait Islander Origin
As for birthplace, the first stage of this analysis compares the PES and Census responses of those for whom Aboriginal/Torres Strait Islander origin was stated
in the Census while the second stage of the analysis considers the response in the PES for persons where the question was not answered in the Census.
Comparison of Census and PES responses for Aboriginal/Torres Strait Islander origin
The percentage of persons whose Census and PES responses agree appears to be very high at 99.53%. It can be seen in Table 15 below that this is due
largely to the high proportion of the population that were classified as being of neither Aboriginal or Torres Strait Islander origin in both the Census and the PES.
The level of agreement was much lower for the other two response categories.
TABLE 15: Responses in the Census and PES for Aboriginal/Torres Strait Islander origin
Of those people who said they were of Aboriginal origin in the Census, 85% said they were of Aboriginal origin in the PES, with most of those who gave a
different response indicating that there were of neither Aboriginal nor Torres Strait Islander origin in the PES. Only 50% of those who said they were of Torres
Strait Islander origin in the Census said they were of Torres Strait Islander origin in the PES, with the majority of those who gave a different response saying
they were of neither Aboriginal or Torres Strait Islander origin in the PES.
Analysis of PES responses for Aboriginal/Torres Strait Islander origin for cases of non-response in the Census
The level of non-response to this question in the Census was 2.98%. Table 16 shows the responses given in the PES by persons who did not respond in the Census.
TABLE 16: PES responses for cases of non-response to Aboriginal/Torres Strait Islander origin in the Census
A higher than expected proportion of people who did not state a response in the Census identified themselves as being of Aboriginal or Torres Strait Islander
origin in the PES.
This analysis of Census and PES responses for variables common to the two collections has a number of implications for the quality of Census data and for
Census form design.
The responses given for age appear to be subject to a large number of small errors, although as most analysis is performed using five year age groups the
implications for the quality of data may not be great. Non-response appeared to be higher for people in older age groups.
The imputation system resulted in many people identified in the PES as being aged between 15 and 19 years (the group most likely not to respond to this
question) being given an age of younger than 15 years. There is also evidence in this analysis of the Census imputation process under-estimating ages
across all broad age groups.
Response to marital status appeared to be reasonably sensitive to variation between the two collections, in particular responses of 'separated' and
'divorced'. The imputation was consistent with PES responses in two thirds of all cases, although the rate was much lower than this for 'separated', 'divorced'
and 'widowed' imputed values.
There was a reasonably high correspondence between responses in the Census and PES to the question on birthplace. Most discrepancies that occurred
involved Australia, Yugoslavia or 'Other' birthplaces. The presence of countries in the list on the Census form appeared to lower non-response for people born
in those countries.
While the overall level of agreement for responses to the question on Aboriginal or Torres Strait Islander origin appeared to be good, this did not apply for
the response categories of 'Aboriginal' or 'Torres Strait Islander' when taken separately. In particular, only half of the Census 'Torres Strait Islander' responses
were borne out by the PES with 40% of the corresponding PES responses being 'Other' (non-indigenous). For about 85% of Census responses of Aboriginal,
the PES responses were consistent. For almost all of the remainder, the PES response was 'Other' (non-indigenous). On the other hand, about 89% of PES
'Aboriginal' responses were consistent with Census responses with most of the remainder being recorded as 'Torres Strait Islander' in the Census.
These documents will be presented in a new window.
This page last updated 19 April 2007