Estimated dwelling stock methodology

Estimated dwelling stock are based on adjusted counts from the 2021 Census of Population and Housing (the Census), updated with quarterly estimates of dwelling additions and removals. Further information on each component can be found below.

This release includes estimates of dwelling stock, additions and removals for the June 2016 quarter to the June 2022 quarter. 

Development funding for these statistics was provided through the National Housing and Homelessness Agreement, contributing to indicator (a) under the Data Improvement Plan. Development funding ended in 2021-22. 

Estimated dwelling stock and its components of change (additions and removals) includes all dwellings as defined by the Functional Classification of Buildings (FCB). This defines a dwelling as a permanent and fixed structure intended for long-term residential use, including cooking and bathing facilities. This differs from the definition of a dwelling used by the Census (a structure which is intended to have people live in it, and which is habitable on Census night), which can include non-permanent structures such as caravans, houseboats and tents. 

Dwelling stock estimates exclude Non-Private dwellings (NPDs) as defined in the Census (establishments which provide communal and short-term accommodation e.g. hospitals, hotels), which are not classified as dwellings in the FCB. Dwelling stock estimates include both Private and Public/Government owned dwellings.

The components of estimated dwelling stock change (additions and removals) classify dwellings according to the original intended function of the building at the time of construction. The Census classifies dwellings based on the building’s use on Census night. Building use can change over time (i.e. from a short-term holiday apartment to a long-term residential dwelling). Changes in building use that occurred between Census 2016 and 2021 have been added into the components of dwelling stock change.

Dwelling stock and its components of change includes all dwellings in Australia excluding Other Territories, as defined by the Australia Statistical Geography Standard (ASGS) Edition 3. Estimates for Other Territories are not provided due to lack of data availability. 

Estimates are prepared at the Statistical Area Level 2, according to the ASGS.

Dwelling stock estimates are classified by type of building according to the Functional Classification of Buildings (FCB). Dwelling stock estimates are available for the following building types: 

• Houses - detached buildings consisting of one dwelling unit, including transportable houses and detached granny flats
• Townhouses - dwellings attached in some structural way to one or more other dwellings, with their own private grounds and no other dwelling above or below
• Apartments - blocks of dwellings that don't have their own private grounds and usually share a common entrance, foyer or stairwell, including attached granny flats
• Total - includes dwellings in non-residential buildings (e.g. caretaker/manager's residence and house/flat attached to a shop) 

In the Census, dwellings are classified according to Dwelling Structure (STRD), which largely aligns with the FCB. There are some STRD classes which do not have a corresponding class in the FCB:

• STRD 91 (Caravan) and STRD 92 (Cabin, houseboat) are included in dwelling stock estimates in the houses category, where they have been identified as fixed structures used for long-term residential purposes. These have been identified based on the Census location of dwelling (DLOD) variable (those in retirement villages (DLOD 4) or manufactured home estates (DLOD 3) are in scope of dwelling stock estimates) and information from ABS field officers. Houseboats are excluded from dwelling stock estimates. Caravans and cabins that are used for short-term accommodation (i.e. in holiday parks) or are mobile (i.e. a caravan where someone usually lives but is moved regularly between sites) are not in scope of dwelling stock estimates. 
• STRD 93 (Improvised home, tent, sleepers out) is excluded from dwelling stock estimates. 
• STRD 94 (House or flat attached to a shop, office, etc.) is included in dwelling stock estimates in the total type of building category only. 
• STRD && (Not stated) has been prorated to STRD and DLOD by meshblock. 

Additions to stock are based on dwelling completions from the Building Activity Survey (BACS), which provides quarterly dwelling completions at state level. Dwelling completions include construction of new dwellings and dwellings created as a result of alterations/additions to existing buildings.

BACS is a quarterly sample survey which uses Building Approvals (BAPS), a monthly administrative collection, as the selection frame. BACS collects information (including the completion date) quarterly from builders until the building job reaches completion or is abandoned. All building approvals with over nine dwellings or valued over $5m are selected in the BACS sample. See Building Activity, Australia methodology for further information.

Small area dwelling completions are compiled from: 

1. Directly collected dwelling completions (building approvals selected in the BACS sample), which are allocated to the SA2 from the building approval and allocated to the quarter in which they were completed as reported by the BACS respondent. Approximately 40% of dwellings approved are selected in the BACS sample (8% of houses, 45% of townhouses and 97% of apartments).
2. Modelled dwelling completions (the remaining building approvals not selected in the BACS sample), which are allocated to the SA2 from the building approval and the probability of completion each quarter contributes to the small area estimate.

The small area completion estimates are calibrated to the state level BACS completion estimates. The small area estimates may then be adjusted: 

• to include other additions to dwelling stock such as change in building use (i.e. a short-term holiday apartment to a long-term residential dwelling)
• to account for timing differences between Census and building activity data (i.e. a staged building job will be completed in building activity once all stages are completed, whereas some stages may have been completed before the Census)
• based on coherence with Census data

The probability of completion for the modelled dwelling completions is estimated using a log-logistic accelerated failure time model with random effects. The following predictor variables available in the building approval data were selected based on the strength of their relationship to the outcome variable (i.e. dwelling completion): 

• State
• Type of building
• Value of approval
• Approval date in 2021 or later (these approvals have experienced higher than usual completion times)

This method produces completion probabilities for each quarter that we expect the dwelling could be completed in (i.e. a dwelling may have a 20% probability of completion in the quarter it was approved, a 30% probability of completion the following quarter etc), which are summed by quarter to create the small area estimate of dwellings completed. 

The errors associated with the modelled estimates for small areas fall into four categories: sampling error, non-sampling error, modelling error, and prediction error. 

Since the estimates for building activity are based on a sample of approved building jobs, they are subject to sampling error; that is, they may differ from the figures that would have been obtained if information for all approved jobs for the relevant period had been included in the survey. Relative standard errors (RSEs) are available in the Building Activity, Australia publication as a measure of the sampling error in building activity statistics. Non-sampling error refers to inaccuracies that may occur in the source of building approval information, imperfections in reporting by respondents, and errors made in the coding and processing of data. 

Modelling error is introduced by model misspecification. This can occur when the choice of model is incorrect, a key predictor variable is left out or an inappropriate predictor variable is included. Therefore, the variables chosen in the models may result in incorrect modelled estimates for certain small areas, particularly small areas that do not follow the typical associations between the available predictor variables and the dwelling completions being modelled. The chosen model has been tested against a range of possible alternative models; however, it is the most suitable model subject to available data at the time. 

A strong model does not guarantee statistically accurate modelled estimates. Prediction error is a measure of the statistical accuracy of the predictions made to produce the modelled small area estimates. The relative root mean squared error (RRMSE) provides an indication of the deviation of the modelled estimate from the true value. The RRMSE is primarily a measure of prediction error, but in its calculation it also inherits some aspects of modelling and sampling error. As a general rule of thumb, estimates with RRMSEs less than 25% are considered reliable for most purposes, estimates with RRMSEs between 25% and 50% should be used with caution and estimates with RRMSEs greater than 50% are considered too unreliable for general use. 

The table below shows the median RRMSEs of the quarterly SA2 dwelling completion estimates, by the size of the estimate. Estimates with higher values (i.e. more dwellings completed) generally have lower RRMSEs. Estimates with very low values (<11 dwellings completed) should preferably be analysed at a broader level (e.g. across multiple quarters or small areas), as these larger aggregated estimates will generally have lower RRMSEs.

Estimates of dwelling stock removals are compiled from:

• Dwelling demolitions – dwellings demolished following a demolition approval (in all states and territories except SA) or electricity abolishment notification (in SA only)
• Unplanned stock losses – dwelling losses resulting from natural disasters

A dwelling demolition is defined as the complete and intentional dismantling of a dwelling, such that none of the structure remains on site. 

Estimates of dwelling demolitions are modelled based on demolition approvals from local councils or other approving authorities (in all states and territories except South Australia) or electricity abolishment notifications (in South Australia only).

Demolitions approvals are sourced from the Building Approvals collection, which capture dwellings approved by a local council or other approving authority to be fully demolished. Within South Australia, approval to demolish a dwelling is only required where the dwelling falls within a historic zone or heritage area. Demolition estimates in South Australia are based on electricity abolishments data sourced from SA Power Networks (SAPN). Abolishment refers to the permanent disconnection of power supply to a premises, a process that is required before dwelling demolition can proceed.

The probability of a demolition going ahead following approval or electricity abolishment notification (in South Australia only) is modelled using state-specific models based on the distribution of approval or electricity abolishment notification to demolition time frames. The model is based on a sample of demolition outcomes, which were observed using aerial imagery from Nearmap Australia Pty Ltd. This provides an approximation of the demolition date, although the frequency of Nearmap image captures will affect the accuracy of the observed demolition date.

This method produces probabilities for each quarter that we expect the dwelling could be demolished in (i.e. a dwelling may have a 20% probability of demolition in the quarter it was approved, a 30% probability of demolition the following quarter etc), which are summed by quarter to create the small area estimate of dwelling demolitions.

The estimates of dwelling demolitions are subject to sampling error; that is, they may differ from the figures that would have been obtained if all demolition approvals and electricity abolishments had been observed using aerial imagery. Non-sampling error may also arise due to inadequacies in the source of building approval or electricity abolishment information. The estimates are also subject to modelling error and prediction error as outlined in the 'additions to dwelling stock' section above, however measures of these errors are not available for dwelling demolitions.

Unplanned stock losses refer to dwellings that are unintentionally and permanently destroyed. This includes dwellings destroyed in natural disasters such as bushfires, floods and cyclones. These types of removals are not accounted for by the Building Approvals collection in a timely way (a decision to seek demolition or building approval is often well after the disaster has occurred). Temporary removals from stock where dwellings may be uninhabitable for a period of time following a natural disaster are not in scope.

Estimates of unplanned stock losses have been compiled based on currently available data from the National Recovery and Resilience Agency (NRRA), which covers the 2019-2020 bushfire season only. This data includes total counts of dwellings destroyed by LGA across the bushfire season. These counts were allocated to SA2s based on the proportion of exposed dwellings in each mesh block within an LGA. This was calculated by multiplying the proportion of the area that was affected by bushfires (based on bushfire boundaries sourced from the NRRA) by the 2016 Census dwelling count. This was then multiplied by the number of destroyed dwellings in the LGA. The estimate of dwellings destroyed in each mesh block was then aggregated to the corresponding SA2. The counts were manually allocated to the appropriate quarters using various news sources. 

A collection of the number of unplanned stock losses is not currently available in a nationally consistent and timely manner. The ABS is continuing to investigate alternatives to estimate unplanned stock losses where direct counts are not available (i.e. based on the location of dwellings and hazard exposure information).

The ABS collects statistical information under the authority of the Census and Statistics Act, 1905. This requires that statistical output shall not be published or disseminated in a manner that is likely to enable the identification of a particular person or organisation.

To guard against identification or disclosure of confidential information, sensitive data cells may be subject to small random adjustments.

The files under the 'Data downloads' heading contain data for small geographic areas in Excel and CSV format (contained within .zip file), as so:

Building Activity, Australia

Building Approvals, Australia

Functional Classification of Buildings

National, state and territory population

Regional population

2021 Census