Homelessness data snapshot: Factors linked to stay length in shelters

• Copyright

  © His Majesty the King in Right of Canada, as represented by the Minister of Housing and Infrastructure,  2026.

  T94-60/9-2026E-PDF
  ISBN 978-0-662-74865-6

On this page

• Purpose
• Data sources
• Methods
• Results
• Conclusion
• Annex A: List of participating communities
• Annex B: Methodology
  • Dependent variable
  • Independent variables
  • Housing affordability ratio and decomposition
  • Regression technique and model specifications
• For more information

Purpose

Figure 1. Summary of Factors Linked to Stay in Shelters

Across Canada, people are spending more time in shelters than they did in the past, and a growing share of shelter users are experiencing chronic homelessness (see the National Shelter Study and the Analysis on Chronic Homelessness). While these trends are well documented, there has been limited analysis of how broader, community‑level conditions contribute to longer shelter stays. A better understanding of this relationship could help governments and service providers to plan more effectively and target their investments.

While individual factors focus on personal circumstances, structural factors look to bigger issues that affect many people. Some examples include rising housing costs, less housing availability, a lack of work opportunities, and low amounts of government financial assistance. This paper analyzes how key structural factors such as social assistance levels, rent costs, vacancy rates, and unemployment rates affect the length of emergency shelter stays.

Much of the existing research on structural drivers of homelessness comes from the United States and shows a strong relationship between higher rents and higher rates of homelessness.^{Footnote 1Footnote 2Footnote 3Footnote 4} Canadian research has reached similar conclusions, finding links between housing costs, shelter capacity, and homelessness.^{Footnote 5} Some recent Canadian work has also examined how labour market conditions and income supports are related to shelter admissions.^{Footnote 6} However, there has been less focus on how changes in these broader conditions affect the length of time people remain in shelters. This analysis helps to fill that gap.

Data sources

To examine shelter stays, this analysis used data on shelter users between 2017 to 2024 from communities across Canada (see Annex A). The data were collected by service providers using the Homeless Individuals and Families Information System (HIFIS) as well as through data-sharing partnerships with communities that use alternative Homelessness Information Management systems. Each person’s shelter use was tracked over a one-year period, allowing us to calculate the total number of days spent in shelter and the average length of stay for each community.

Income plays a key role in helping people exit homelessness, especially their ability to afford housing. Because detailed income information is not available for individual shelter users, this analysis uses average social assistance benefit levels as an indicator of income. These benefit levels were compared to average rents to better understand how affordable housing is for people relying on social assistance. Provincial social assistance data from 2017 to 2024 were taken from Maytree’s Welfare in Canada Reports.

Other data highlight why this comparison is important. Census results show that nearly 85% of people staying in shelters have incomes below Canada’s low income threshold, and shelter users are far more likely to rely on social assistance than people living in private housing.^{Footnote 7} Research also shows that social assistance benefits are generally not high enough to lift people out of poverty across most household types.^{Footnote 8}

Housing availability and affordability also affect how quickly someone can leave a shelter. In markets where housing is scarce or expensive, it can take longer for people to find and secure housing. These conditions were measured using median rents and rental vacancy rates from the Canadian Mortgage and Housing Corporation’s annual Rental Market Survey.

Finally, employment opportunities can improve income and housing options. Local labour market conditions were captured using annual unemployment rates, sourced from Statistics Canada, to reflect how job availability changes over time within communities.

Methods

We used a statistical analysis to understand how different structural factors are related to the length of time people stay in shelters. This approach allows us to look at the effect of each factor on its own, while taking into account differences between communities and changes over time.

The analysis also adjusts for broader trends that affect all communities, such as pandemic‑related disruptions and rising costs due to inflation. To ensure the results were robust, we tested several versions of the analysis.^{Footnote 9} See Annex B for technical details.

Results

When rent affordability, unemployment rate or vacancy changes within a community, do shelter stays change?

As housing affordability worsens within a community over time, people tend to stay in shelters longer. For example, even a $100 rent increase leads to longer shelter stays, but how much longer depends on the assistance amount. For a community where social assistance is an average of $1000 for individuals, it means about 10 extra days in shelter.

Other factors, such as local unemployment rates and rental vacancy rates, were not linked to the length of shelter stays. This finding aligns with previous research from both Canada and the United States.

Conclusion

Housing affordability affects local shelter systems in multiple ways. Less affordable rents are not only associated with higher shelter admissions^{Footnote 10} and a greater prevalence of homelessness^{Footnote 1Footnote 2Footnote 3Footnote 4Footnote 11} but this study also finds that they are linked to longer shelter stays.

Higher inflows into shelters and longer lengths of stay significantly increase overall demand for shelter services, resulting in more individuals being sheltered on any given night. As a result, shelter occupancy rises placing additional pressure on systems that are already under strain.

Annex A: List of participating communities

List of participating communities from 2017 to 2024

Community	Year
	2017	2018	2019	2020	2021	2022	2023	2024
Abbotsford	-	-	-	-	x	-	x	x
Barrie-Simcoe	-	-	-	x	x	x	x	x
Brant-Brantford	x	-	-	-	x	x	x	x
Chilliwack	x	-	-	-	-	-	-	-
Durham Region	x	-	-	-	x	x	x	x
Fredericton	-	-	-	-	x	x	x	x
Halifax	x	x	x	x	x	x	x	x
Hamilton	x	x	x	x	x	x	x	x
Kamloops	x	-	-	x	x	x	x	x
Kelowna	x	x	-	x	x	x	x	x
Kingston	-	-	-	-	x	-	x	x
London	-	-	x	x	x	x	x	x
Moncton	x	-	-	x	-	x	x	x
Nanaimo	x	x	x	x	x	x	x	x
Niagara Region	-	-	-	-	-	x	x	x
Ottawa	x	x	x	x	x	x	x	x
Peterborough	-	-	-	-	-	x	x	x
Regina	x	-	-	-	-	-	-	-
Saint John	-	-	-	-	-	x	x	x
Sudbury	x	-	-	-	x	-	x	x
Thunder Bay	-	x	-	-	-	x	x	x
Toronto	x	x	x	x	x	x	x	x
Vancouver	x	x	x	-	-	x	x	x
Victoria	x	x	-	-	-	-	-	x
Waterloo Region	x	x	x	x	-	x	x	x
Wellington-Guelph	-	-	x	x	-	x	x	x
Windsor	-	-	-	-	-	x	x	x
Winnipeg	x	-	-	-	-	x	x	x

Annex B: Methodology (technical details)

Dependent variable

The dependent variable was the average cumulative number of days spent in shelter per community. Data on shelter stay length were collected by service providers using the Homeless Individuals and Families Information System (HIFIS) and through data sharing partnerships with jurisdictions that use other homelessness information management systems.

Communities were included if they had more than 84% shelter system coverage for three consecutive years. This ensured that there was sufficient coverage to accurately capture shelter use over time. These communities represent multiple provinces and a range of city sizes. A full list of the communities included in this analysis can be found in Annex A.

For each calendar year from 2017 to 2024, annual shelter use was calculated using the total number of shelter days accumulated over a 365 day observation window. Clients were included if they resided in the selected communities and had at least one shelter stay that occurred within, or crossed, the calendar year boundaries (1 January to 31 December).

For each client, the observation window was defined as the 365 days preceding the client’s last recorded book out date in the year of interest. All shelter days recorded within that window were summed to produce a client level total. These client level totals were then summarised at the community level to produce the annual community measure used in the analysis (Table 2).

Table 1: Number of communities included in the analysis in each year and the range of average stay lengths 

Year	Number of communities	Average cumulative length of stay over 365 days
		Minimum	Maximum
2017	17	19	100
2018	10	26	103
2019	9	31	117
2020	12	34	137
2021	15	30	137
2022	22	35	155
2023	25	33	169
2024	26	39	188

Independent variables

Table 2 summarises the variables included in the regression models. These measures capture key characteristics of the housing market, social safety net, and labour market in each jurisdiction. Variables were selected primarily based on the availability of annual, community level data and their use in previous research on structural factors associated with homelessness.

Housing availability and affordability. Housing availability and affordability reflect the challenges individuals may face when attempting to re‑enter the housing market. Less available and more expensive housing markets may contribute to longer shelter stays. These conditions were represented using median rents and primary rental market vacancy rates, sourced from the annual Canadian Mortgage and Housing Corporation’s annual Rental Market Survey.

Social assistance. Low income can limit a person’s ability to secure independent housing. Census analysis indicates that 84.7% of shelter residents aged 15 and older reported after tax incomes below the Low Income Measure (LIM) threshold. Shelter residents were also substantially more likely to receive social assistance (54.1%) than individuals living in private dwellings (4.6%).^{Footnote 7} Additional evidence indicates that social assistance benefit levels fall below the official poverty line for 98% of household types across provinces.^{Footnote 8} While exact income information for shelter users is not available in these data, average social assistance benefit levels provide a useful proxy for income constraints. Provincial social assistance data (2017–2024) were sourced from Statistics Canada Labour Force Survey.

Table 2: Independent variables included in analysis, their source, jurisdictional level and reporting frequency 

Data point	Frequency	Jurisdictional level	Source
Historical Primary Rental Market Vacancy Rate (%)	Annual	Municipal	CMHC Rental Market Survey
Historical Primary Rental Market Median Rent	Annual	Municipal	CMHC Rental Market Survey
Monthly social assistance (average)	Annual	Provincial	Maytree Welfare in Canada Report
Year	Annual	N/A	N/A
Unemployment Rate	Annual	Municipal	Statistics Canada. Unemployment rates used by Employment Insurance program, three-month moving average, seasonally adjusted

Housing affordability ratio and decomposition

To capture the relationship between income supports and local housing costs, a single housing affordability measure was used: the ratio of average monthly social assistance to the historical primary rental market median rent (range: 0.34 to 0.99). This ratio reflects the adequacy of income support relative to local rental costs; lower values indicate lower affordability.

To assess how each component contributes to differences in affordability between communities, a log decomposition analysis was conducted:

Var(log(ratio)) = Var(log(average social assistance)) + Var(log(median rent) − 2Cov(log(average social assistance), log(median rent)) 

Results of the decomposition were as follows:

• Share due to social assistance: 37% of observed variation between communities
• Share due to rent: 117% of observed variation between communities
• Share due to the relationship between rent and social assistance reduces observed dispersion by 57% (because higher rent communities also tend to have higher benefit levels)

Overall, the decomposition indicates that variation in rent across communities is the dominant driver of differences in affordability. Social assistance levels also contribute, but to a more moderate extent. The positive relationship between rents and benefit levels partially offsets rent effects but does not eliminate the affordability gap between higher rent and lower rent communities.

Regression technique and model specifications

A fixed effects panel regression was used to account for the structure of the data (repeated observations of communities over time). Year fixed effects were included to control for changes affecting all communities each year (e.g., pandemic related disruptions and inflation).

Because the panel is unbalanced, multiple model specifications were estimated to confirm that results were not driven by the unbalanced structure:

1. Full unbalanced panel.
2. Panel restricted to communities with five or more years of shelter data.
3. Balanced panel restricted to communities with eight years of shelter data.

Regression results across specifications are summarised in Table 3.

Table 3: Selected panel regression results

Variable	(1) Full unbalanced panel	(2) Panel with 5 or more years of data	(3) Balanced panel (8 years of data only)
Vacancy Rate	0.686	1.164	1.454
Standard Error (SE)	(1.52)	(1.58)	(0.785)
Ratio of monthly average social assistance to median rent	106.295 **	100.842 **	135.043 **
Standard Error (SE)	(34.407)	(35.018)	(46.962)
Unemployment rate	-0.707	-0.819	-0.665
Standard Error (SE)	(1.02)	(1.22)	(3.070)
Year fixed effects	Yes	Yes	Yes
Number of observations	119	96	48
Group variable	Community	Community	Community
R² within	0.6802	0.6858	0.6447
R² between	0.2738	0.3510	0.6705
R² overall	0.4241	0.4482	0.6037
** The result is statistically significant.

Across specifications, the association between housing affordability (rent to assistance ratio) and shelter stay length remained positive and statistically significant, including when the sample was restricted to communities with five or more years of data and when limited to the balanced panel.

In the full unbalanced panel, a 0.1 increase in the rent to assistance ratio (i.e., rent becoming less affordable relative to assistance) was associated with an increase of approximately 10.6 days in the average shelter stay within communities over time. In contrast, unemployment rate and vacancy rate were not statistically associated with shelter stay length.

Overall, the full, unbalanced model explains 68% of the year to year change in shelter stay length within jurisdictions (within R²).

For more information

Find out more about homelessness research on Reaching Home: Canada's Homelessness Strategy.

If you have any questions about this report, contact us.