ABS Address Register, Users' Guide

Describes the maintenance of residential addresses, dwelling structure and machine learning used for Census and household survey frames.

Released
14/12/2020

Overview

The ABS Address Register is an up-to-date, comprehensive list of all known physical addresses within Australia.

The key information it contains is:

  • address text
  • address use (e.g. residential, commercial, etc.)
  • a latitude and longitude coordinate known as a “geocode”
  • dwelling structure for residential addresses (e.g. separate house, semi-detached, etc.).

The Register does not collect or store any information about people or households.

It is a key piece of statistical infrastructure that supports:

  • mail out of Census of Population and Housing (Census) and Household survey materials
  • more effective sample selection methods for household surveys
  • more efficient field operations
  • additional internal data quality assurance
  • linking of datasets
  • enhanced address matching through the creation of indexes
  • finer geospatial analysis at the unit/address level.

How it is used

The Address Register is used to produce the Address Register Common Frame (Common Frame), a point in time snapshot of the addresses on the Register.

 The Common Frame is used to produce the:

  • Census Mail Out Frame
    A dataset of addresses used to support the mail out of Census materials to households across Australia. For the 2021 Census
    , the Address Register expects to aid the delivery of Census materials to 85% of households either by post or by a Census field officer.
  • Household Survey Frames
    Datasets of addresses which are in scope for ABS household surveys including the Labour Force Survey and Special Social Surveys. Household surveys use these frames in their sample selection processes.
  • Counts of Addresses by Address Type, Dwelling Structure and Special Dwelling
    Summary statistics for internal use.

Other uses:

How the data is collected

Administrative data

The Address Register is based on the Geocoded National Address File (G-NAF) © Geoscape Australia Copyright and Disclaimer Notice. The Register differs from G-NAF as it contains information about address use and dwelling structure which is important for household surveys and the Census.

Other administrative datasets used to update the Register include:

  • Australian Electoral Commission data
  • ABS Building Approvals and Demolitions data 
  • Australia Post's Postal Address File supplied to the ABS by Intech Solutions
  • residential property transaction data supplied to the ABS by CoreLogic.
     

Survey feedback

ABS field officers report back information about addresses (e.g. address use and dwelling structure) during household surveys and the Census to improve the Register. This does not include information about people that may live at an address.

Desktop canvassing

Desktop canvassers analyse information about addresses using online resources including aerial imagery, property listings, and state/territory government planning and mapping tools to update the Register.

Aerial imagery

Aerial images sourced from Nearmap Australia Pty Ltd are used as an input for desktop canvassing and the Automated Image Recognition (AIR) machine learning model.

Scope

The Address Register includes all known physical addresses within Australia (including Other Territories). The Register does not include non-physical addresses such as post office boxes or email addresses.

Address use

Addresses on the Register belong to one or more address use categories:

  • residential
  • commercial
  • under construction
  • vacant land
  • Special Dwelling (see below)
  • other address use types (e.g. carparks, parkland, etc.).
     

Special Dwellings

There are two types of Special Dwellings:

Non-Private Dwellings (NPD)

Non-Private Dwellings provide communal accommodation or care such as hotel/motels, hospitals and nursing homes. There are currently 24 Non-Private Dwelling types.

Private Dwelling Establishments (PDE)

Private Dwelling Establishments consist of a number of addresses co-located or grouped together, often on one land parcel. All dwellings within the boundaries of the establishment (including any manager or caretaker residences) are considered part of the establishment. There are four types of Private Dwelling Establishments:

  • retirement villages
  • manufactured home estates
  • caravan parks and camping grounds
  • marinas.

To maintain confidentiality, the Address Register does not store information about Special Dwellings which are sensitive (e.g. women's domestic violence refuges). These establishments have no publicly available information about them. They are listed on the Register as regular residential addresses.
 

Dwelling Register of Aboriginal and Torres Strait Islander Communities (DRATSIC)

The DRATSIC includes all dwellings within all:

  • discrete Aboriginal and Torres Strait Islander communities
  • Aboriginal and Torres Strait Islander outstations
  • Aboriginal and Torres Strait Islander town camps across Australia.


The types of dwellings on the DRATSIC are:

  • permanent private dwellings
  • non-private dwellings.


The DRATSIC does not include:

  • non-residential structures within communities
  • pastoral stations
  • cultural/seasonal camping grounds
  • Indigenous Community Places with no permanent dwellings
  • regional centres.

Data items

The key information stored for each address on the Address Register is:

Address text components

Special Dwelling data items

DRATSIC data items

Privacy and confidentiality

The ABS is committed to upholding the privacy, confidentiality and security of the personal information it collects. The ABS has strong legislative protections that safeguard the secrecy of an individual’s information and a proud history of maintaining community trust in the way it collects, uses, discloses and stores personal information.

The Address Register does not collect or store any information about people or households.

Access to the Register is restricted to ABS staff who have a demonstrated business case for using the data. This is managed via an access control list.

Quality

Coverage

The Address Register aims to minimise under coverage and over coverage of residential addresses across Australia. Total Common Frame counts are compared against data sources such as the Census Post Enumeration Survey (PES) to provide a measure of coverage.

Under coverage

Under coverage occurs when residential addresses do not exist on the Register, or appear on the Register as a different address type when they are actually residential. This can impact on the data quality for household surveys and the Census. During the 2016 Census, under coverage was estimated to be 0.9%.

Over coverage

Over coverage occurs when a residential address on the Register is either not able to be located in the field, or is able to be located but is not in scope for the survey (e.g. an address is listed as residential but is still under construction). This makes data collection for household surveys and the Census more expensive and time consuming. However, it does not affect survey data quality. Using Labour Force Survey data, over coverage is estimated to be 4.9% for the twelve months ending in February 2020. (Note: Estimates up to February 2020 have been included as COVID-19 has made recent address coverage more difficult to evaluate at this point in time).

Timeliness

The Address Register is updated on a quarterly schedule. Overall, there can be a lag of up to six months in the data on the Register.

Factors contributing to the time lag are:

  • a three-month lag between receiving new G-NAF addresses and the release of the next Common Frame. G-NAF data is received every quarter (February, May, August, November). Updates made from G-NAF become available to users of  the Address Register frame in the following quarter. For example, the G-NAF supplied in May will be reflected on the August Common Frame.
  • a two-month lag between receiving new Postal Address File addresses and the next Common Frame. For example, the Postal Address File data received at the end of August will be reflected on the November Common Frame.
  • a three to six-month lag before Building Approvals data becomes visible on the Common Frame. For example, updates to the Register based on Building Approvals for the months of June, July and August will be seen in the November Common Frame.
  • a lack of up to date aerial images in remote areas. Several processes used to update the Register rely on recent aerial images as a source of information. Very remote areas receive infrequent updates, often less than once per year.

Address Register maintenance

The processes used to validate and update addresses each quarter are:

The application of a predictive administrative data model

An administrative data model assigns address use and dwelling structure information to new and existing addresses on the Register. It combines several administrative datasets to classify the addresses.

The application of business rules using administrative data

Pre-defined sets of business rules use administrative data to capture changes in address use from demolition, subdivision, construction and conversion activities (for example from a residential to commercial premise).

Automated Image Recognition

A machine learning image classification model, known as the Automated Image Recognition (AIR) model, is used to rapidly classify addresses into address use categories based on aerial images.

Desktop canvassing

Desktop canvassers manually update such information as address use, dwelling structure and geocodes for new and existing addresses based on their observations. This process is generally reserved for complex addresses which cannot be classified by the data models or business rules.

Quality gates

A comprehensive series of quality gates are used to ensure the quality of the Register. This includes quality gates for new administrative datasets, updates to the Register, and for the production of the Common Frame.

Standards and classifications

Addresses on the Address Register have been classified using the following:

Conceptual Framework

The core unit for the Address Register is a ‘physical location with an address'. A unique Address Register Identifier (ARID) is created and stored against each core unit.

Principal and alias addresses

An ARID can have one or many addresses that accurately describe its location. For example, 10 Smith Street might also be known as 10-12 Smith Street. For the Register, there is a “principal address” for the core unit, while the remaining addresses are labelled as “alias addresses”. All associated principal and alias addresses are linked to the same ARID.

Source address information

The Register stores information about where an address is sourced from, whether the source be G-NAF or ABS collected data (e.g. desktop canvassing or ABS field officer feedback).

Special Dwellings

Private Dwelling Establishments (PDEs) such as retirement villages and manufactured home estates are represented by a Special Dwelling Parent and linked Special Dwelling Children for the individual dwellings within these establishments. Special Dwelling Children are represented directly as core Register units.

Non-Private Dwellings (NPDs) and all other Private Dwelling Establishments (PDE) (that are not Special Dwelling Children) are represented by a Special Dwelling Parent.

Each Special Dwelling Parent is assigned a unique Special Dwelling Identifier (SD ID) and a Special Dwelling Register Number (SDRN) and is linked to its corresponding core address (ARID) on the Address Register.

Some ARIDs might have multiple Special Dwelling Parents linked if they are co-located at the same address. Examples include Nursing Home/Retirement Village, or Accommodation for the Aged (Hostel)/Nursing Home.

Dwelling Register of Aboriginal and Torres Strait Islander Communities (DRATSIC)

Each DRATSIC dwelling unit is assigned a DRATSIC Dwelling Identifier, which is linked to the corresponding core address on the Address Register. DRATSIC dwelling units are grouped together by community, known as an Indigenous Community Place, which is also recognised by a unique identifier (ICP ID).

Automated Image Recognition

AIR is a machine learning model that classifies aerial images of addresses. It is a part of the Address Register maintenance process. AIR results are combined with administrative data to update information about address use.

Benefits of Using AIR

  • Accuracy
    AIR has an accuracy of 96.7% for the three most important address use types (residential, under-construction, and vacant).
  • Improved Efficiency
    AIR is able to classify approximately 600,000 images over a five-day working week compared to 1,000 for a human over the same period. This speed allows the Register staff to more regularly monitor existing addresses and classify new addresses. AIR also enables the Register to allocate resources more appropriately. It is best used to classify large numbers of simple addresses such as separate houses which make up approximately 70% of residential addresses on the Register. This allows the desktop canvassing team to focus on more complex addresses, such as hotels and holiday parks.
  • Improved coverage 
    Research revealed that administrative data is only available for 68% of new addresses. Aerial imagery extends that coverage to 96%.
  • Timeliness
    Administrative data can indicate that an address is residential before the dwelling has been constructed. Recent aerial imagery can confirm whether a dwelling exists at that address. This is useful for updating vacant and under construction addresses.
     

How AIR Works

AIR is built using a machine learning algorithm known as a convolutional neural network. Convolutional neural networks place images into categories based on the combinations of patterns that exist within a given image. AIR classifies images into one of six categories:

  • Residential
  • Under-Construction
  • Vacant Land
  • Road
  • High Density
  • Commercial
     

AIR was initially trained on 6,000 hand classified aerial images sourced from Nearmap Australia Pty Ltd. Each image run through AIR is centred on the geocode (latitude and longitude) of an address.

To achieve high quality results, AIR checks when the aerial image was captured to ensure the results are still relevant. Property size is also used as an indicator for whether an address can be classified by AIR. If a property size is too large, the address is excluded from AIR processing as the dwelling at the address may not be captured in the image. Register staff can also visually confirm classifications made using AIR for quality assurance.

History

The Address Register was created using the April 2015 G-NAF and was validated via the:

  • 2015 National Field Canvassing Exercise
    Field canvassing officers visited 92% of addresses across the country, verifying addresses and capturing relevant information such as address use and dwelling structure.
  • 2016 Census
    Census field officers provided feedback about addresses (e.g. address use and dwelling structure) to improve the Register.

The first Common Frame was produced in 2015 to support the 2016 Census. Quarterly Common Frames have been produced since April 2017. The Census model underwent significant changes in 2016, adopting a digital first strategy. In previous Census years, ABS field officers visited the majority of households across Australia in-person to deliver and collect Census forms. With the creation of the Register, the ABS was able to mail Census materials to approximately 60% of households, greatly increasing efficiency and reducing costs.

For more information on the Address Register, see Contact Us.

Non-Private Dwelling Types

The 24 Non-Private Dwelling types are:

NPD Category    Description
NPD011               Hotels-Motels
NPD012               Bed and breakfast
NPD031               Nurse's Quarter
NPD032               Mining Camps
NPD033               Staff Quarters (not mining camps)
NPD040               Boarding house, private hotel
NPD050               Boarding School
NPD060               Residential College/Hall of Residence
NPD070               Public Hospitals (not psychiatric)
NPD080               Private Hospitals (not psychiatric)
NPD090               Psychiatric hospital or institution
NPD100               Hostel for the disabled
NPD110               Nursing Home
NPD120               Accomodation for the retired or aged (not self contained)
NPD130               Hostel for the homeless, night shelter and refuge
NPD140               Childcare institution
NPD150               Corrective institution for children
NPD160               Other Welfare Institution
NPD170               Prison or corrective institution for adults
NPD180               Immigration detention centre
NPD190               Convent, monastery, etc
NPD200               Youth/backpacker Hostel
NPD210               Ski lodges
NPD220               Other, not elsewhere classified

Glossary

Show all

Abbreviations

Show all

Back to top of the page