From substance use to homelessness or vice versa?

https://doi.org/10.1016/j.socscimed.2015.05.005Get rights and content

Highlights

  • This paper investigates links between homelessness and substance use.

  • We exploit Australian data collected in 4 waves (2 years) for 1325 ‘at risk’ individuals.

  • We show a strong positive association between homelessness and substance use.

  • We provide evidence that this association is unlikely to be causal, in either direction.

  • The exception is a positive effect from risky alcohol use to homelessness.

Abstract

Homelessness is associated with substance use, but whether substance use precedes or follows homelessness is unclear. We investigate the nature of the relationship between homelessness and substance use using data from the unique Australian panel dataset Journeys Home collected in 4 surveys over the period from October 2011 to May 2013. Our data refer to 1325 individuals who were homeless or at risk of becoming homeless. We investigate dynamics in homelessness and substance use over the survey period. We find that the two are closely related: homeless individuals are more likely to be substance users and substance users are more likely to be homeless. These relationships, however, are predominantly driven by observed and unobserved individual characteristics which cause individuals to be both more likely to be homeless and to be substance users. Once we take these personal characteristics into account it seems that homelessness does not affect substance use, although we cannot rule out that alcohol use increases the probability that an individual becomes homeless. These overall relationships also hide some interesting heterogeneity by ‘type’ of homelessness.

Introduction

The prevalence of homelessness is difficult to measure, but a recent Department of Housing and Urban Development snapshot estimate for the United States (US) suggested around 630,000 people were sleeping on the streets or in shelters in January 2012 (HUD, 2012). This estimate would be considerably higher were those ‘doubling up’ with family or friends or in other forms of insecure housing included, and higher still if the number in question referred to people experiencing homelessness within a period of time rather than at a single point in time. For example, Link et al. (1994) estimated that 4.6% of the US population had been homeless at some point over the 5 years between 1986 and 1990. More recently O'Flaherty (2012) has again stressed the importance of thinking about homelessness from a dynamic perspective.

Not only does homelessness deprive individuals and families of a basic human need (Curtis et al., 2013), it is also strongly associated with a wide range of other social problems. For example, levels of substance (ab)use are far higher among the homeless than among the wider population (e.g. Greene et al., 1997, Shinn et al., 1998, Early, 2005, Kemp et al., 2006). Indeed there is a widely-held view that homelessness and substance use are self-reinforcing, i.e. that substance use causes homelessness (e.g. Allgood and Warren, 2003, Early, 2005), and that homelessness causes substance use (e.g. Shinn et al., 1998, Johnson and Chamberlain, 2008), or both (e.g. Johnson et al., 1997, Neale, 2001). Whether a causal link exists between homelessness and substance use, and if so of what magnitude and in which direction(s), are obviously crucial questions for policy makers and service providers designing and delivering interventions in this area.

Unfortunately the lack of consensus in this regard reflects an evidence base which is at best patchy. At the heart of this problem lies a dearth of suitable data on the substance use of representative samples of individuals experiencing homelessness and individuals at risk of homelessness observed over time. This is not an easy population to reach, let alone reach repeatedly. As a result many studies are based on small-scale, ad hoc, cross-section surveys (Scutella and Johnson, 2012). These surveys also tend to be of very specific – often acutely homeless or acutely using – populations such as clients of treatment centres, other service providers, or ‘skid row’ communities in urban centres (e.g. Teesson et al., 2000, Booth et al., 2002). This is also the case, albeit to a lesser extent, for larger cross-sections studied in this literature like the National Survey of Homeless Assistance Providers and Clients (NSHAPC) (see e.g. Early, 2005). Other studies exploit cross-sections enhanced with retrospective information (including the NSHAPC), but again these tend to be of specific provider-based populations (e.g. Allgood and Warren, 2003, Johnson and Chamberlain, 2008). Studies using longitudinal data also tend to be based on small samples of similarly specific populations, often with a very limited time dimension, and with little attempt made to deal with unobservable confounders (e.g. Allgood et al., 1997, Zlotnick et al., 2003, Rice et al., 2005, Kemp et al., 2006). The net result is a body of literature from which it is difficult to draw general conclusions and from which we can learn little regarding causality even within the study populations themselves.

Two more promising studies using longitudinal data, covering homeless and at-risk-of-homelessness individuals, are Shinn et al. (1998) and Fertig and Reingold (2008). Fertig and Reingold (2008) exploit data from the Fragile Families and Child Wellbeing Study tracking around 5000 low-income parents from the birth of their child for a further three years. Once they control for a host of observable individual and contextual factors they find no evidence that mothers with a ‘drug problem’ – those who report that their drinking or drug use interfered with their work or personal relationships over the previous 12 months – are any more likely to be homeless at the one-year follow up interview. Shinn et al. (1998) use survey data on around 250 families requesting shelter accommodation in New York City and a similar number of comparison families drawn from welfare records, with both groups re-interviewed five years later. They also find little evidence of a substance abuse impact on homelessness once other observable factors are controlled. Neither study, however, examines whether homelessness impacts on substance use in a multivariate model, both draw on only small numbers of people experiencing homelessness (from quite particular populations), and both have limited information on substance use (single dummies in each case, with the bar set high to be counted as a substance abuser).

The current paper is the first to examine the dynamic inter-relationships between homelessness and substance use, potentially running in both directions, in a large-scale, broadly–based longitudinal survey, drawing on a population covering individuals experiencing differing degrees of homelessness and comparable individuals not currently homeless but at risk of homelessness, and with richly detailed data on substance use. The data come from the Australian Journeys Home (JH) study, which is unique in its scale, detail and coverage (see Wooden et al., 2012), and has not previously been used to study the links between homelessness and substance use beyond a handful of descriptive tables and brief accompanying discussions in the in-house research reports linked to the release of the first four waves of data (Scutella et al., 2012, Chigavazira et al., 2013, Johnson et al., 2013, Scutella et al., 2013).

Specifically, we address four research questions. First, what are the extent, nature and persistence of substance use among homeless people and those at risk of homelessness in the JH sample? Second, what is the association between substance use and homelessness status in the JH sample? Third, to what extent might these associations be driven by causal relationships from substance use to homelessness, from homelessness to substance use, or both? Fourth, do these relationships vary according to type of homelessness?

In addressing these questions we make a number of specific contributions. We provide the most detailed description of the use of alcohol, cannabis, and other illegal/street drugs among JH respondents during the first 2 years (4 waves) of the survey to date. Second, we demonstrate strong point-in-time associations between substance use and homelessness for this broad-based sample. Third, we show that these associations are predominantly driven by observed and, crucially, unobserved individual characteristics which cause individuals to be both substance users and homeless. Once we take these characteristics into account, and appealing to the arrow of time to infer the direction of any remaining relationship, we conclude that homelessness does not seem to affect substance use in the next 6 months, while only risky alcohol use seems to increase the probability that an individual becomes homeless in the next 6 months. Finally, we show that some substance use behaviours appear to impact heterogeneously on different types of homelessness. The implication is that conclusions regarding the associations between substance use and homelessness are likely to be sensitive to the definition of homelessness used, in particular whether those ‘doubling up’ with friends and family and those in other temporary accommodation are included along with those sleeping rough.

The remainder of the paper is set out as follows. Section 2 provides descriptive information about homelessness and substance use in the JH sample. Section 3 presents our exploratory analysis of the determinants of homelessness and Section 4 the determinants of substance use. Section 5 discusses sensitivity analysis and presents estimates by homelessness type. Section 6 concludes. An online appendix includes further data details and results.

Section snippets

The JH data and variable definitions

JH is a longitudinal dataset with information on a sample of recipients of any income support (i.e. welfare) payment who are either homeless or at-risk of homelessness in Australia (Scutella et al., 2012). The Melbourne Institute ran this survey for which data collection was approved by the Human Ethics Committee of the University of Melbourne. Here we use the first four waves collected between September 2011 and May 2013 focussing on the balanced panel, i.e. the 1325 respondents who were

Determinants of homelessness

In the previous section we presented evidence of a number of associations between homelessness and substance use. However, these associations are point-in-time and unconditional, i.e. they ignore dynamics and they may be driven by differences in individual and contextual characteristics that influence both homelessness and substance use. To further investigate the association between substance use and homelessness we estimate a series of linear models for homelessness, initially separately for

Determinants of substance use

In the same way we explore potential causal effects of homelessness on substance use, estimating linear models separately for each substance, first including only wave dummies as controls, then including observable time-varying and time-invariant controls as in (1), and finally replacing the time-invariant observed controls with individual fixed effects. Our preferred model – the fixed effects model – is given by (2):Sit = α1Hit + α2Hit−1 + α3Zit + θi + φt + μit,where Sit denotes substance use

Sensitivity analysis and extensions

First consider sensitivity to the precise definitions of our substance use variables. We explore three specific changes: varying the threshold at which alcohol consumption is defined as risky, replacing the cannabis use variable with a dummy for daily cannabis use, and replacing the other illegal/street drug variable with a dummy for weekly use. Lowering or raising the threshold at which alcohol consumption is defined as risky does not impact qualitatively on our conclusion of a possible causal

Summary and conclusions

This paper exploits unique longitudinal data for a large and broadly-based sample of homeless and at-risk-of-homelessness individuals to examine the dynamics of homelessness, substance use and, using standard panel data methods, the associations between substance use and homelessness. In doing so it makes a number of significant contributions to a mostly descriptive literature bedevilled by a dearth of suitable data to examine these issues in anything other than small and often very specialised

Acknowledgements

This paper uses unit record data from Journeys Home: Longitudinal Study of Factors Affecting Housing Stability (Journeys Home). The study was initiated and is funded by the Australian Government Department of Social Services (DSS). The Department of Employment has provided information for use in Journeys Home and it is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the authors

References (27)

  • A.R. Fertig et al.

    The characteristics and causes of homelessness among at risk families with children in twenty American cities

    Soc. Serv. Rev.

    (2008)
  • J.M. Greene et al.

    Substance use among runaway and homeless youth in three national samples

    Am. J. Public Health

    (1997)
  • HUD

    The 2012 Point-in-time Estimates of Homelessness

    (2012)
  • Cited by (87)

    • Do early episodes of depression and anxiety make homelessness more likely?

      2022, Journal of Economic Behavior and Organization
      Citation Excerpt :

      The literature has also investigated the role of job loss in triggering mental health issues (for example, Salm, 2009; Marcus, 2013; Bünnings et al., 2017) and homelessness episodes (Metraux et al. 2018 for a review). More relevant to young people, the use of illegal drugs has been linked to both mental health (Van Ours and Williams, 2012, Moschion and Powdthavee, 2018) and to homelessness (McVicar et al., 2015, 2019). Likewise, experiences of trauma, and sexual and physical violence are associated with subsequent mental health conditions and homelessness.

    • Dynamics of drug overdose in the 20th and 21st centuries: The exponential curve was not inevitable, and continued increases are preventable

      2022, International Journal of Drug Policy
      Citation Excerpt :

      In Geoffrey Rose's classic paper “Sick Individuals and Sick Populations” (Rose, 1985), Rose notes that the risk factors for who develops certain health conditions within a population versus across two populations may be distinct. Indeed, there are many risk factors for substance use disorder within any population, including adverse childhood and other stressful life events (Stein et al., 2017), poverty and housing insecurity (McVicar, Moschion, & van Ours, 2015; Walker & Druss, 2017), drug use history (Muhuri, Gfroerer & Davies, 2013), and psychiatric comorbidity (Fenton et al., 2012). Why some individuals develop substance use disorders and others do not, within a community, can partially be explained by some of these well-established risk factors.

    View all citing articles on Scopus
    View full text