Australian Bureau of Statistics

Rate the ABS website
ABS Home > Statistics > By Catalogue Number
2050.0.55.002 - Position Paper - ABS Review of Counting the Homeless Methodology, Aug 2011  
Latest ISSUE Released at 11:30 AM (CANBERRA TIME) 02/08/2011  First Issue
   Page tools: Print Print Page Print all pages in this productPrint All RSS Feed RSS Bookmark and Share Search this Product  
Contents >> Contents >> Complexities in estimating homelessness


It is not easy to estimate the number of people who are homeless in Australia. Few countries in the world attempt the complex task of comprehensively estimating the number of homeless people in their population. The homeless population is small compared to the entire population, is spread over a wide geographical area and is often difficult to find during enumeration (and in fact some may not wish to be found). Consequently, people who are homeless are difficult to capture in a statistical collection.

Homelessness can result from such a diversity of reasons that anyone in society can potentially experience homelessness at some stage in their lives. This compounds the difficulties in measuring the homeless population because it can be difficult to identify unique characteristics of this population in order to identify them, amongst all people in the community, at the time that they are experiencing homelessness. Even if the homeless population is included in a statistically representative data collection, classifying them as homeless in output remains a challenge because not only are people's experiences of homelessness varied and complex in nature, often they may not identify themselves as being homeless.

For both enumeration and classification reasons, the homeless population is probably one of the hardest populations to enumerate and understand in a statistically representative way.

Being homeless may mean that some people are not captured at all in datasets used to count people generally. This includes the Census of Population and Housing, which is the most comprehensive enumeration of the Australian population. Nor will all homeless people be necessarily counted in datasets that count only those homeless people accessing particular homeless services, or those obtaining government benefits. Due to the difficulties in identification and collection, no existing measures of homelessness are precise.

There are many different aspects of homelessness that are of interest to understand the number of homeless people and their characteristics. For example, the need to count everyone in the Australian population, including those who are homeless, is important so that services and funding can be appropriately distributed across the Australian population. It is also important to appropriately classify the homeless population to provide understanding of the characteristics of those who are amongst the most disadvantaged in Australia. The reasons for understanding homelessness may affect the kind of estimate which is needed. Different dimensions include, but are not limited to, prevalence or point-in-time estimates, incidence or numbers experiencing homelessness over time, characteristics of those who are homeless, transitions into and out of homelessness, and the length of typical experiences of homelessness. Some of these are discussed below.



A fundamental estimate of homelessness is prevalence. Prevalence is an estimate of how many people experienced homelessness at a particular point-in-time. A prevalence estimate should ensure that each person is included only once in the estimate if they were homeless at a particular point in time. An accurate measure of the prevalence of homelessness allows society to judge the scale of the problem. If prevalence measures can be estimated on a consistent, comparable basis and at regular intervals, then trends and the direction of change can be determined. This allows monitoring of the numbers of those who were homeless, and can be used to identify if interventions or policies have been successful. It allows society to hold itself and governments accountable for some outcomes at this broad level.

As importantly, if policy and program action is to target preventing, or ameliorating the circumstances of homelessness, knowing the locations of the homeless, and their characteristics, is necessary for effective targeting. Such knowledge also allows monitoring of the outcomes of programs to identify what interventions are successful. Ideally, fine geographic level prevalence measures allow consideration of where homeless people are located for place based targeting of services and other interventions. The characteristics of the homeless population, such as sex, age, whether of Aboriginal or Torres Strait Islander origin, or the identification of subpopulations such as those who are in primary or secondary homeless situations are also valuable delineations of a point-in-time measure for interventions tailored to client needs.


A second measure of homelessness is an estimate of the number of people experiencing at least one period of homelessness over a given period of time, for example, over a 12 month period. This is a flow measure and is sometimes referred to as an incidence measure. Incidence measures have value in informing service provision by showing the potential demand for services over a given period.

Incidence measures show all experiences of homelessness over a period, and may include multiple incidences of homelessness for some individuals. It is difficult to find a collection vehicle to count these phenomena in real time. Counts of those who access services is one option, however these incidence measures are constrained to the services available, and to their population of service users. People's recall of their experiences is another approach, but recall accuracy can be an issue, especially for repeat periods of homelessness.

Individuals' experiences of homelessness

Data on the typical length of an experience of homelessness, along with the type of homelessness experienced and where the person stayed while homeless, are all important. An estimate of the number of experiences of homelessness per person over a given period of time also increases understanding of movements into, and out of, homelessness. Critical to the analysis of such experience data is the ability to identify sub-populations and the characteristics of persons with particular experiences. This enables services to be developed and targeted to certain types of homeless experiences.

Obtaining extra information on the events that both trigger how a person becomes homeless, and how they move out of it, and potentially back again, are also important. This helps to develop programs to intervene before a person reaches the point where they become homeless (or homeless again) and to intervene to support their movement out from a state of homelessness. The characteristics of people, including identifying the types of people who may be at risk, and the events that might trigger homelessness, are important data items to make this information useful for policy and program planning.


Potential data sources for estimating the different aspects of homelessness that have been identified include: the Census of Population and Housing; administrative data from service provider collections; the homelessness flags on Centrelink records; ABS household surveys such as the General Social Survey, Personal Safety Survey, etc; and longitudinal data sources including the 5% Statistical Longitudinal Census Dataset and the longitudinal study Journeys home - Longitudinal Study of Factors Affecting Housing Stability. The potential for, and limitations of, these sources to enhance statistical understanding of homelessness are discussed below.

In addition to data sources that can present a statistically representative picture of the homeless population and allow comparison to the total population, individual qualitative studies (including case studies) are important to understand homelessness in more depth. They can provide the histories of those who have been homeless, and attempt to understand, for some individuals, what interventions work. They are also important to understand the key factors that statistically representative surveys should focus on. Both quantitative and qualitative research are important in understanding the broad spectrum of issues relating to homelessness.

Census of Population and Housing

The Census of Population and Housing (conducted by the ABS on a five-yearly basis) provides the best opportunity for a point-in-time or prevalence estimate of homelessness that can be used in trend estimation.

The power of the Census in estimating homelessness is that the Census aims to count all persons in Australia on Census night (with the exception of foreign diplomats and their families). It includes those people who are in conventional private dwellings but also those in non-private dwellings or those who were not in a dwelling at all on Census night. As a result, aside from the challenges of enumerating them on Census night, persons who are homeless are in scope of being counted in the Census. The Census also collects detailed information about people such as those who have a need for assistance, those born overseas, and Aboriginal or Torres Strait Islander people. With data available at a very fine geographic level, the Census also provides the potential for estimates of homelessness to be calculated for most geographic areas.

Some submissions to the review noted that the net undercount in the 2006 Census was estimated by the ABS at 549,486 people, and that homeless people are likely to be over represented in that net undercount. This could imply a significant underestimate of the homeless population that is based on the Census results. Any such undercount would impact on the reviewed estimates of homelessness, other than for those people enumerated in SAAP properties, i.e. it would impact on the accuracy of the estimate of 17,000 or so homeless people in boarding houses, the 2,000 in other non-private dwellings, the 8,000 in improvised dwellings and the 20,000 in other private dwellings.

The quality of the Census data is further enhanced by using information collected in a post-enumeration survey (PES) to obtain estimates of the net undercount in the Census. The PES is conducted immediately following the Census. While the PES collects information representative of the vast majority of Australians, it is not designed to estimate the undercount of persons who may be homeless (as it does not cover all people, such as those who live in special dwellings such as boarding houses, or those who are not living in private dwellings at the time of the PES). It does, however, provide information about the characteristics of people who may have been missed in the Census. It will include some people who were homeless on Census night but were not homeless during the PES, or those who were staying in a private dwelling on Census night such as those people staying with other households.

There were 6,276 people enumerated in the 2006 PES and for whom a Census record was not found (ABS 2007a, Table 4.1). Some 97% of these respondents to the PES were usual residents of their PES address. Only 3% were visitors at the time of the PES. Of the 4,796 PES respondents who thought that they had been counted in the Census three weeks earlier, about three-quarters thought that they would have been counted at their PES address i.e. at home. They were not visitors to that address at either the time of the Census nor at the time of the PES. For these people with a common usual address at both the Census and the PES, the reasons for them being missed during the Census are not known. Common reasons include simple error on the part of householders completing the form (e.g. assuming the person is being enumerated elsewhere because they were away from home on Census night, or being accidentally left off the Census form). If they were in a specialist homeless services dwelling at the time of the Census and have since returned home, they would have been included in the SAAP component of homelessness in the Census dataset. If they had been in a boarding house at the time of the Census and were missed in a boarding house on Census night, their usual address elsewhere in Australia (their PES address) would have meant they would need to be excluded from any homeless estimate.

The other 25% of this group of 4,796 PES respondents who thought they had been counted (but in fact wasn't) nominated another address where they thought they were enumerated in the Census. About 50% of these people (546) nominated a Census night dwelling address that was missed by Census collectors i.e. they were not people who chose not to participate, nor were they in dwellings that the Census collectors could not make contact with. They were in dwellings that the Census collector simply missed and their omission from the Census enumeration has no impact on homelessness estimates.

While 97% of PES respondents reported their PES address as their usual address, and most of those reported that it was also their Census night address, there may be some people enumerated in the PES who were not counted in the Census because they were temporarily absent and homeless at the time of the Census, but had returned home in the three weeks since the Census. Such circumstances may include youth or people escaping domestic violence and staying temporarily with other households on Census night but not being recorded on the Census form for that household.

And as the PES does not approach non-private dwellings (nor people sleeping out) it does not generate direct estimates of undercount for people in those circumstances at the time of the PES. However, the final undercount estimates are weighted to account for the entire population, including those people in non-private dwellings and also includes people who were not in dwellings.

While there is so much value from the Census in creating prevalence estimates of homelessness, it does have some limitations. It is not possible to ascertain homelessness from a direct Census question, so instead, to use Census data to identify the size and characteristics of the homeless population, it is necessary to create decision rules to infer whether persons enumerated are, on balance, likely to be homeless. This means that the Census can only be used to create estimates of the number of persons who may be homeless, not measure this characteristic directly.

In the methodologies used to estimate homelessness, the data items available from the Census may not support the assumptions being made when interpreting the data, for three possible reasons:

  • they do not adequately characterise homelessness (for example employment or education status); and/or
  • the underlying questions have not been designed for the purpose of homeless identification (for example, the usual address question); and/or
  • because the characteristics may be poorly reported (for example, a person reported a usual address when they do not in fact have one).

While none or not all these reasons apply to each characteristic or assumption, overall they mean that the measurement of homelessness from the Census may overestimate or underestimate the number of people classified in the Census dataset as homeless on Census night.

Underestimation is likely to be greater for population groups, such as for Aboriginal or Torres Strait Islander Australians, which have experienced relatively high rates of undercount in previous Censuses. In addition, Indigenous Australians may report a usual address that is culturally associated with a place rather than with an adequate shelter (see Chapter 7 of the Discussion Paper and the Key Issues section of this Position Paper). Undercount in the Census is also more likely for people (including Indigenous Australians) staying in dwellings, such as public housing, without any legal right to occupy the premises. The completed Census form for such dwellings may simply show the tenants, and not any of their friends or family who may be staying on Census night.

Youth are also likely to be underestimated in the homeless population. For youth, such as 'couch surfers', to be classified as homeless in the Census, reporting 'no usual address' on the Census form is critically important. For youth who are homeless and staying with another family, this question may be incorrectly completed because the member of the family they are staying with may report the youth's previous address on the Census form as their 'usual' address. This may be because they do not know that the youth can not return to this address. Alternatively some youth may not admit to having no usual address as they do not want the stigma attached to being 'homeless'.

Underestimation is also likely for victims of domestic violence who, at the time of the Census, may assume they still have, and therefore report, the usual address from which they have fled. For others, they may not report themselves on the Census form out of fear that their location will be identified. However, people who are escaping domestic and/or family violence will be identified as homeless if they are staying in crisis accommodation, or in boarding houses if they report that dwelling as their usual residence, while a significant proportion staying temporarily with another household or in boarding houses may not be classified as homeless because they report a usual address elsewhere.

Overall, estimating homelessness from the Census is through identifying those who are most likely to be homeless on Census night based on a number of characteristics. These estimates cannot include those people who were not enumerated in the Census, because, other than the PES, there is no estimate of the numbers who were not enumerated.

The ABS is focussed on improving enumeration in the 2011 Census for a number of key population groups including Aboriginal and Torres Strait Islander Australians, migrants, fly-in/fly-out workers, the homeless and those who will not be home on Census night. This is not only to ensure that the Census comprehensively covers the whole Australian population, but also to improve estimation of key population groups such as the homeless or Indigenous people. The ABS will achieve this through a range of special enumeration strategies. Seeking improved enumeration is the focus for new or improved methods for the future, and planned actions are noted and recommendations for further improvement are made in Future Directions section.

Several countries undertake rough sleeper census counts, crisis accommodation census counts and/or utilise administrative data to capture these aspects of homelessness on their Census night. However, no other country currently attempts a prevalence measure across all aspects of homelessness. Some countries only undertake a Census every 10 years, and/or do not ask a usual address question, therefore limiting the usefulness of the Census to undertake further homelessness analysis. Professors Chamberlain and MacKenzie were groundbreaking in researching whether the Australian Census could provide insight into estimating the broader homeless population on Census night.

Through the methodological review, the ABS has concluded that the Counting the Homeless estimates did not satisfy the requirements for measuring prevalence or trends over time, but that through testing and refining the estimation methods, the ABS has confidence the Census can provide trend analysis of the size and characteristics of the homeless population on Census night. It is not yet clear how close such Census based estimates are likely to be to the true measure of homelessness in the population at that point in time, and ABS will work further with stakeholders to refine the measures for that purpose. However, applying a consistent methodology helps to understand change over time in the numbers of those who are homeless, even if it cannot estimate the exact size of the population.

SAAP National Data Collection (until 30 June 2011)

Until the end of June 2011, the Supported Accommodation Assistance Program National Data Collection (SAAP NDC) was the main source of data on the provision of services through the SAAP program. Until recently, three components to this collection existed: the Client Collection, the Administrative Data Collection and the Demand for Accommodation Collection. SAAP funding covered both supported accommodation and related support services, so data from the collection covered the number of services provided to people experiencing homelessness or being at risk of experiencing homelessness, and these were presented as support periods and as accommodation support periods. Data from the Client collection can be used to generate estimates of the number of people experiencing at least one period of SAAP support for a given period of time (see the Key Issue section of this Position Paper for more information).

The collection provided further detail about the characteristics of clients (and accompanying children) who received SAAP services. The data enhanced understanding of the characteristics of people who sought homeless services and who gain access to these services. Some data items include the type of support received, reasons for clients seeking assistance, circumstances of clients before and after support and the collection can show changes in support provided over time (AIHW, 2011).

It may be difficult to extrapolate service provider data such as that from the SAAP NDC to make statements about the underlying homeless population. For example, if a service reported a doubling of accommodation services provided within a time period, this may mean that the actual homeless population had doubled or that the population accessing accommodation has doubled through increased accommodation being provided, or through an increased knowledge of the services available. Instead, the value of service provider data is in reporting about service usage and about the characteristics of those accessing the services. It is important to remember that the characteristics of those seeking and/or accessing services may differ from the characteristics of those who did not.

SAAP NDC data provided support period data, or flow data, not point-in-time data so it is not useful to match or augment the Census point-in-time prevalence estimates. Separate SAAP data were collected and compiled in relation to accommodation provided on Census night for use is assessing the quality of Census data on people enumerated in SAAP dwellings on Census night.

The SAAP client collection provides data about the number of active accommodation support periods at a point-in-time (eg. on Census night), as well as the characteristics of those accessing the services. The data were usually collected in May and September. However, this collection was moved to August in Census years to provide a point-in-time estimate of the number of active accommodation support periods and has been used to verify Census SAAP estimates as mentioned in the previous paragraph and as discussed in Discussion Paper: Methodological Review of Counting the Homeless, 2006 (ABS cat. no. 2050.0.55.001). These data are useful alongside the Census point in time estimates for providing a richer picture of homelessness for these service users.

Specialist Homeless Services (SHS)

The SAAP collection has been replaced by a new collection - Specialist Homelessness Services (SHS) collection that commenced on 1 July 2011. AIHW will compile the data, the first of which will become available in 2012. This Specialist Homelessness Services (SHS) data collection will provide data about the pathways people take in and out of homelessness and the kinds of work homelessness agencies do. It will be able to identify individual clients as well as support periods and children will be counted as individual clients. In addition, family information will be more accurate. Information about previous episodes of homelessness and people turned away from homelessness agencies will also be available. The data will be able to provide snapshots of homelessness at a given point in time, which was not previously available with the past datasets (AIHW 2011).

Centrelink 'vulnerability to homelessness flag'

Centrelink have included homelessness 'flags' in their system which help customer service officers provide appropriate services to people experiencing homelessness or who are at risk of homelessness. Data produced using these indicators may provide useful information about those who are identified by Centrelink as homeless. It may be possible to analyse this population against other data items held by Centrelink to provide a picture of the characteristics of the homeless population in receipt of Centrelink benefits (FaHCSIA 2010).

The flags have been implemented to inform Centrelink staff that the client needs active follow up to ensure that they are receiving the support they need and are able to meet any obligations arising from their income support payment. It will also be used by Centrelink to change business practices to better meet the needs of such vulnerable clients. The flag is not designed to be a measure of all people on the Centrelink database who are either experiencing or are at risk of homelessness, but rather it shows those who have been identified as being in this group through client/service interaction. Data extracted using this indicator may be biased towards areas where Centrelink staff have been better trained or are more proactive in identifying and using the indicator. In addition, the indicator will not cover the entire homeless population as there will be some homeless people who will not be on the Centrelink database and some Centrelink customers who are not in enough contact with Centrelink to be identified as relevant for the flag to be applied.

While the flag may not provide an estimate of the absolute number of people who might be homeless, it may be able to provide an indication of movement in homelessness, as well as the characteristics of those people who are homeless or at risk of homelessness.

ABS 2010 General Social Survey

The ABS 2010 General Social Survey included a new homelessness module. The survey interviews one randomly selected adult per household and collects information about the respondent's socio-demographics including income, wealth, social participation measures, employment, education, problems accessing services etc. The new homelessness module will be able to identify previous experiences of homelessness and provide insight into the homelessness experience of the population. This module included whether the person had a period of time without a permanent place to live, and if so, whether they had been accommodated in a range of circumstances (eg. night shelter, with friends or relatives, slept rough etc). The survey also collects data on what led to the homeless circumstance, and the frequency with which thay have experienced homelessness. For the most recent experience of homelessness, data are collected on when that homeless experience occurred, for how long, and whether services were approached for assistance, what assistance was provided (if applicable) or why services were not approached.

The GSS only collects information from people who are in private dwellings. The survey does not approach people who live in non-private dwellings such as boarding houses, or those who are not in dwellings at all. Therefore it cannot inform on current homeless experiences.

The data about previous experiences of homelessness can be cross-classified with all of the other social capital variables collected in the GSS. This includes their income, wealth, feelings of safety, experiences of violence, contact with friends and relatives, problems accessing services etc.

GSS data are expected to inform on the flows through homeless periods in the 12 months prior to the survey, in the two years prior and in the five years prior. ABS will be publishing data from the 2010 General Social Survey in late September 2011.

The next General Social Survey, to be run in 2014, will also include an enhanced homelessness module to enable comparisons with 2010 of previous experiences of homelessness.

Other ABS Surveys

The ABS proposes to consider the inclusion of a GSS-like homelessness module in other future ABS surveys, as appropriate, such as the Survey of Disability, Ageing and Carers, the Survey of Income and Housing and/or the Household Expenditure Survey.

The ABS will also investigate the development of a culturally appropriate module on the previous experiences of homelessness for the 2014 National Aboriginal and Torres Strait Islander Social Survey.

Personal Safety Survey 2012

The ABS is testing questions for potential inclusion in the Personal Safety Survey (PSS) 2012. These could cover information about a person's housing arrangements the last time they separated from a violent current partner and their housing arrangements at the end of their last violent previous partner relationship. The ABS is proposing to seek information from respondents who have experienced current partner violence to establish, whether they have ever separated from their violent current partner and had to leave their home, and if so, where they went the last time they separated. The ABS is also seeking to establish from respondents who experienced violence from a previous partner, when they left their last violent previous partner, whether they had to leave their home, and if so, where they went when the relationship finally ended. For example whether they stayed with a friend or relative, slept rough, stayed in a refuge or shelter, stayed in temporary accommodation eg. motel etc or elsewhere. If they went to multiple places, we ask them for the place in which they spent the most time.

While not a complete picture of where people went every time they separated during all relationships, if the testing is successful, this will provide an indication of what accommodation was used by people the last time they separated from their violent partner/s.

Journeys home: Longitudinal Study of Factors Affecting Housing Stability

As part of the National Homelessness Research Agenda, FaHCSIA is funding Journeys Home: Longitudinal Study of Factors Affecting Housing Stability, the first large-scale longitudinal study following the lives of 1,550 Australians who are homeless or who may be vulnerable to homelessness. Participants will be interviewed every six months over two years and the results of the study will assist in understanding the various factors associated with homelessness and housing stability.

A 5% Statistical Longitudinal Census Dataset (SLCD)

The ABS is planning to create a Statistical Longitudinal Census Dataset (SLCD) by bringing together data from the 2006 Census with data from the 2011 Census and future Censuses to build a picture of how society moves through various changes: which groups are affected by different types of change and in what way. The 2006 SLCD dataset and the 2011 Census dataset will be brought together using a statistical method referred to as 'probabilistic record linkage'. This involves bringing together data from the two datasets without using names and addresses but by using a number of characteristics common to both datasets such as age, sex, geographic region and country of birth (for more information see Census Data Enhancement Project: An Update, Oct 2010, ABS cat. no. 2062.0).

The ABS will investigate using the 5% SLCD to undertake longitudinal analysis of the circumstances of those who have been identified as likely to be homeless. The circumstances of people identified as likely to be homeless on the 2011 SLCD can then be compared with their circumstances in 2006, and into the future it should be possible to report on repeat periods of homelessness and long term outcomes as seen in the SLCD. It will also be possible to compare these results, for those likely to be homeless, with the rest of the population. As outlined in the Census Data Enhancement Project paper referenced above, the ABS may enhance the 5% SLCD further by bringing it together with other non-ABS datasets (without using name and address) which would provide additional information for analysis (such as housing or health data).


The ABS will use any data sources that are, or will become, available to check and/or refine the prevalence measure derived from the Census, or to help to understand and quantify mis-estimation of any sub-populations. This includes validation against the estimates generated from the current rules as well as to validate any new estimates generated from either a refinement of the derivation rules or an augmentation using new data. The additional datasets, as outlined earlier, will be examined over time to aid testing the decision rules applied to the Census. These datasets include the new specialist homelessness collection, the Journeys Home longitudinal survey, the General Social Survey (and future surveys using the homelessness module), the Centrelink homelessness/at risk of homelessness flag, the data generated by confronting the Census/Centrelink data sets, as well as sector derived datasets or other research that is relevant and emerges through responses to the Discussion Paper or through the ongoing work of the new Homelessness Statistical Reference Group.

The Position Paper explores a number of issues with measuring homelessness. There is a particular focus on issues with the use of Census data to create a prevalence measure of homelessness, and on the complexity that is homelessness that impinges on the use of Census data to report on homelessness in many circumstances. Some of the issues covered include:
  • Complexities of estimating homelessness
  • The definition of homelessness
  • Flow measures of homelessness.

This edition of this Position Paper covers other key issues with measuring homelessness. Some of the issues covered include:
  • Aboriginal and Torres Strait Islander peoples
  • Youth
  • Domestic violence
  • The marginally housed
  • Overcrowding
  • Construction workers
  • Travellers
  • Recently arrived migrants and culturally and linguistically diverse populations
  • Other non-private dwellings.

Previous PageNext Page

Bookmark and Share. Opens in a new window

Commonwealth of Australia 2014

Unless otherwise noted, content on this website is licensed under a Creative Commons Attribution 2.5 Australia Licence together with any terms, conditions and exclusions as set out in the website Copyright notice. For permission to do anything beyond the scope of this licence and copyright terms contact us.