This document was added or updated on 12/02/2016.

SAMPLE METHODOLOGY

SCOPE AND COVERAGE
LINKING METHODOLOGY
WEIGHTING, BENCHMARKING AND ESTIMATION
SOURCES OF ERROR
DATA CONSISTENCY



SCOPE AND COVERAGE

The ACLD is a random 5% sample of persons enumerated in Australia on Census Night, 8 August 2006 which has been linked using statistical techniques to records from the 2011 Census, conducted on 9 August 2011. The Census covers all areas in Australia and includes persons living in both private and non-private dwellings but excluding:

• diplomatic personnel of overseas governments
• persons who expected to be usually resident in Australia for less than six months
• Australian residents overseas on Census Night

Overseas visitors are excluded for the 2006 ACLD sample. Visitors within Australia to private and non-private dwellings on Census Night are included.

For more information on the scope and coverage of the Census:

• 2903.0 - How Australia Takes a Census, 2006
• 2903.0 - How Australia Takes a Census, 2011

LINKING METHODOLOGY

Data from the 2006 ACLD sample and the 2011 Census were brought together using data linkage techniques. The method involved linking without the use of name and address, as this information is destroyed at the end of each Census processing cycle.

Data linkage is typically undertaken using probabilistic and/or deterministic methods, both of which were used in forming the ACLD:

• Probabilistic: linkage is based on the level of overall agreement on a set of variables common to the two datasets. This approach allows links to be assigned in spite of missing or inconsistent information, providing there is enough agreement on other variables to offset any disagreement.
  
• Deterministic: linkage involves assigning record pairs across two datasets that match exactly or closely on common variables. This type of linkage is most applicable where the records from different sources consistently report sufficient information and can be an efficient process for conducting linkage.

Variables on the 2006 and 2011 Census files used for linking include:

• Age
• Sex
• Date of birth
• Indigenous status
• Birthplace
• Year of arrival
• Marital status
• Level of qualification
• Field of qualification
• Highest year of school completed
• Occupation
• Religion
• Language spoken
• Mother's age
• Mother's day and month of birth
• Father's age
• Father's day and month of birth
• Meshblock
• Statistical Areas 1, 2 and 4

A number of linkage passes were conducted based on different combinations of variables to ensure each record had the highest possible chance of being linked. At the end of the linkage process, 800,759 (82%) of the 979,661 sample records from 2006 were linked to a 2011 Census record.

There were two reasons why some records from the 2006 Census were not linked to a 2011 record:

1. Records belonging to the same individual were present at both time points but these records failed to be linked because they contained missing or inconsistent information.
2. The person had no record in the 2011 Census.

For detailed information on the linking methodology and an assessment of its quality see Australian Census Longitudinal Dataset, Methodology and Quality Assessment (cat. no. 2080.5).

Variables relating to migrants from the Department of Social Services' Settlement Database have been included into the ACLD. These have been taken from an existing linkage between the Australian Census and Migrants Integrated Dataset. For information on the linking methodology of Settlement Database variables see Australian Census and Migrants Integrated Dataset Linking Methodology (cat. no. 3417.0.55.001).


WEIGHTING, BENCHMARKING AND ESTIMATION

Weighting

Weighting is the process of adjusting a sample to infer results for the relevant population. To do this, a 'weight' is allocated to each sample unit - in this case, persons. The weight can be considered an indication of how many people in the relevant population are represented by each person in the sample. Weights were created for linked records in the ACLD to enable longitudinal population estimates to be produced. Cross-sectional population estimates for 2006 and 2011 are available from each Census.

The ACLD began as a random sample of 5% of the Australian population in 2006. As such, each person in the sample should represent about 20 people in the population. Between Censuses, however, the in scope population changes as people die or move overseas. In addition, Census net undercount and data quality can affect the capacity to link equivalent records across waves. The ACLD weights benchmarked the linked records to the population that was in scope of both the 2006 and 2011 Censuses. The weights were based on four components: the design weight, undercoverage adjustment, missed link adjustment and population benchmarking.

The original population benchmark was the 2011 Estimated Resident Population (ERP). The 2011 ERP was chosen over the 2006 ERP as the baseline population as it is more recent. The 2011 ERP was then adjusted so as to exclude people who were not in Australia in 2006 as depicted below.




Weights were benchmarked to the following population groups:

• state/territory by age (ten year groups) by sex by mobility (interstate arrivals benchmarked separately)
• Indigenous status by state/territory
  

At 12 February 2016 a new weight was applied to the ACLD file to better account for overseas departures and arrivals between 2006 and 2011. Users who have analysed the ACLD prior to 12 February 2016 may notice changes to estimates produced with the revised weight. Estimates of population groups will be different with the total weighted population estimate being 19.5 million compared to 18.6 million on the old weight. Proportions are expected to only show small differences when previous tables are compared.

The weights have a mean value of 24 and range between 17 and 103. Higher weights are associated with people of Aboriginal and Torres Strait Islander origin and people who moved interstate between 2006 and 2011.

Estimation

Estimates of population groups are obtained by summing the weights of persons with the characteristic(s) of interest.SOURCES OF ERROR

All reasonable attempts have been taken to ensure the accuracy of the results of the longitudinal dataset. Nevertheless potential sources of error including sampling, linking and census quality error should be kept in mind when interpreting the results.

Sampling Error

Sampling error occurs because only a small proportion of the total population is used to produce estimates that represent the whole population. Sampling error refers to the fact that for a given sample size, each sample will produce different results, which will usually not be equal to the population value. There are two common ways of reducing sampling error - increasing sample size and utilising an appropriate selection method (for example, multi-stage sampling would be appropriate for household surveys). Given the large sample size for the ACLD (1 in 20 persons), and simple random selection, sampling error is minimal.

Linking Accuracy

False links can occur during the linkage process as even when a record pair matches on all or most linking fields, it may not actually belong to the same individual. While the methodology is designed to ensure that the vast majority of links are true, some are nevertheless false. The nature of the process used for the ACLD linkage means that while the links obtained are to a high degree of accuracy, some false links may be present within the ACLD dataset. There is an estimated 5% -10% false link rate in the ACLD.

For further detail on the accuracy of the linkage, see Australian Census Longitudinal Dataset, Methodology and Quality Assessment (cat. no. 2080.5).

Managing Census Quality

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 procedures. There are four principle sources of error in Census data: respondent error, processing error, partial response and undercount. Quality management of the Census program aims to reduce error as much as possible, and to provide a measure of the remaining error to data users, to allow them to use the data in an informed way.

Respondent error

For most households in Australia, the Census is self-enumerated. This means that householders are required to complete the Census form themselves, rather than having the help of a Census collector. The Census form may be completed by one household member on behalf of others. Error can be introduced if the respondent does not understand the question, or does not know the correct information about other household members. Self-enumeration carries the risk that wrong answers could be given, either intentionally or unintentionally.

Processing Error

Much of the data on the Census form is recorded using automatic processes, such as scanning, Intelligent Character Recognition and other automatic processes. Quality assurance procedures are used during Census processing to ensure processing errors are kept at an acceptable level. Sample checking is undertaken during coding operations, and corrections are made where necessary.

Partial Response

When completing their Census form, some people do not answer all the questions which apply to them. While questions of a sensitive nature are generally excluded from the Census, all topics have a level of non-response. This can be measured and is generally low. In those instances where a householder fails to answer a question, a 'not stated' code is allocated during processing, with the exception of non-response to age, sex, marital status and place of usual residence. These variables are needed for population estimates, so they are imputed using other information on the Census form, as well as information from the previous Census.

Undercount

The goal of the Census is to obtain a complete measure of the number and characteristics of people in Australia on Census Night and their dwellings, but it is inevitable that a small number 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, thus the effect when both factors are taken into account is a net undercount.

For more detail see Managing Census Quality.
DATA CONSISTENCY

A small percentage of linked records have inconsistent data, such as a different country of birth at the two time points or an age inconsistency of more than one year (when the expected five year difference is accounted for). Inconsistencies may be due to:

• reporting error - information for the same individual was reported differently in 2006 and 2011
• processing error - the value of a data item was inaccurately assigned or imputed during processing
• false link - the record pair does not belong to the same individual

In most analysis, the effect of inconsistent information has a very small impact. Characteristics from either the 2006 or 2011 data can be used in tables and some exploration of consistency over time will assist in drawing appropriate conclusions.

No data editing was applied to the file beyond that which had already taken place during the relevant Census processing period. A set of consistency flags has been included on the ACLD file so that inconsistent data may be observed, quantified or excluded from calculations. Consistency flags, located in the Longitudinal group of data items, have been created for Census variables that would not be expected to change over time or have unlikely transitions over time. These are as follows:

• Age
• Birthplace of Person
• Birthplace of Male Parent
• Birthplace of Female Parent
• Sex
• Year of Arrival
• Number of Children Ever Born
• Registered Marital Status
• Highest Year of School Completed
• Level of Highest Non-School Qualification
• Country of Birth of Spouse or Partner
• Age of Spouse or Partner
• Indigenous Status

There are numerous ways to define consistency. The consistency flags have fine level categories to allow users flexibility in using their own definition of consistent or inconsistent. For example where one Census has 'not stated' for the year of arrival data item, a user can decide whether the record should be considered consistent or not. The same applies to where the response for one Census is 'not applicable'. The labels attached to each category suggesting consistency or inconsistency will assist the user in determining which records are consistent or inconsistent for their needs.

See also Longitudinal Data Items in the Data Items chapter.

Characteristic	Proportion of linked records with inconsistent data between 2006 and 2011
	%
Age (within 1 year)	2.4
Sex	0.1
Birthplace of Person	2.1
Birthplace of Female Parent	4.0
Birthplace of Male Parent	4.4
Year of Arrival	16.5
Indigenous Status (either newly identified or previously identified as Aboriginal and/or Torres Strait Islander)	0.5
Registered Marital Status	0.7
Highest Year of School Completed	6.3
Level of Highest Non-School Qualification	14.9
Country of Birth of Spouse or Partner	2.7
Age of Spouse or Partner	7.9