Entries and exits from homelessness: A dynamic analysis of the relationship between structural conditions and individual characteristics

This report examines the relationship between structural factors, individual characteristics and homelessness. Our interest in the interaction of structural conditions and individual characteristics gives rise to two secondary research questions. First, do structural factors such as housing and labour market conditions, as well as area-level poverty, matter for those individuals vulnerable to homelessness? Second, do structural factors affect those with particular individual risk factors more than others? The questions were answered by analysing an individual's probability of being homeless, the probability of the housed entering homelessness, and the probability that homeless individuals will exit homelessness. In the first two chapters we provide background material while Chapter 3 describes our approach. We rely on economic choice theory and a housing demand and supply framework to set up the empirical approach. The research approach estimates three models that include the individuals' static homeless state, as well as the dynamics of individuals' homelessness through an examination of entry into and exit out of homelessness. The static model includes all observations in the Journeys Home (JH) dataset to assess the probability that an individual will be homeless at each interview. To estimate the probability of entry into homelessness, we identify all persons who are classified as housed and estimate their probability of entering into homelessness in the next six months (i.e. being classified as homeless at the next interview). To analyse the probability of exiting homelessness, we focus on those persons who are classified as homeless, and estimate their probability of becoming housed at the next wave. A random effects logit model is employed to perform the estimations in each model. In each model we present the mean marginal effects of each of the covariates to assess both the statistical significance and the magnitude of effects. In Chapter 4 we outline our data sources and the definition of homelessness. To estimate the contribution of structural factors and individual characteristics requires both micro-level (individual) longitudinal data and area-level data that capture the conditions of social structures such as the housing and labour markets of areas. In the past, micro-level longitudinal data has not been available but this changed with Journeys Home, a longitudinal survey of Centrelink customers who were homeless, at risk of homelessness, or who have high propensity to be homeless (vulnerable to homelessness). The Journeys Home data is ideal for examining the interactions between structural conditions and individual characteristics as it includes detailed information on individuals' characteristics and housing circumstances over time, as well as biographical information prior to the survey. It also covers a representative and sizeable number of geographic areas, with the initial sample clustered across 36 areas drawn from all states and territories, and follow-up interviews attempted even when initial sample members move to areas outside of these initial clusters. Since the sample is designed to be representative of those living in insecure housing circumstances, we are analysing whether individuals belonging to a vulnerable group (due to either personal characteristics or structural factors) are at higher of lower risk of homelessness as compared to others living in insecure housing circumstances. We draw on area-level data from the 2011 Census to establish our housing market conditions measure. The median rent of private rentals is the key measure, which typically reflects the level of housing demand relative to its supply in an area, and is commonly used as an indicator of the tightness of housing markets. A number of other measures such as the demand and supply of low cost housing are also tested. The indicator of local labour market conditions is the regional unemployment rate sourced from the ABS monthly Regional Labour Force Statistics (ABS 2014). As our housing market measure is time invariant, we average the monthly unemployment rates over the Journeys Home Survey period (over two-and-a-half years) to ensure the consistency between the two measures. Housing and labour market data is provided at Statistical Area Level 4 (SA4). It is questionable whether SA4s are the appropriate classification to use when representing capital city residents exposure to housing and labour market conditions. Thus, each of the three models is estimated using two different spatial unit definitions of the area based variables. In our preferred specifications, SA4s within the greater capital city areas have been merged, with unemployment rates and median weekly rents measured on a city-wide basis where relevant. Estimations are also performed using area-level measures in a finer spatial unit classification where the SA4 spatial unit is retained across all Australian regions, including greater capital city metropolitan areas. In Chapter 5 we present the results. For the static model we found that men, older people (45 years plus), those with low educational attainment, the unemployed (or those outside the labour market) are at higher risk of homelessness. So too are individuals who have experienced recent violence or who have recently been incarcerated. The static model also finds that individuals experiencing episodes of primary homelessness prior to JH were also at greater risk of homelessness. Surprisingly, regular drug use was not significantly associated with homelessness, nor was the absence of parenting during childhood, or involvement in the child protection system. Factors correlated with these behavioural and biographical characteristics could be responsible for the elevated rates of homelessness associated with these groups. Similarly, those with diagnosed mental health issues (bi-polar or schizophrenia) are at less risk of homelessness than those without a similar condition. We speculate that diagnosis of these conditions makes delivery of support services more likely. There is confirmation from the estimates that people who were married, had dependent children, or who had better social support, are less likely to be associated with homelessness. After controlling for personal characteristics and risky behaviour, we find housing markets matter, but the evidence on the effects of labour markets is mixed. The analysis of homeless status during JH provides an indication of the overall effects of structural and individual risk factors on homelessness, but the picture provided by the static analysis is far from complete. Factors that may affect an individual's likelihood of entry into and exit from homelessness may be different, and if so a more nuanced perspective on the likely effectiveness of different forms of policy intervention is required. Thus, we estimate models of the probability of entry (for the housed) and probability of exit (for the homeless) separately. The results of the entry and exit models are also presented in Chapter 5. Our entry model provides further confirmation that vulnerable males are less likely to sustain secure housing than females. We also find that the presence of children lowers the chances of becoming homeless, regardless of relationship status. Those with resident children are 2.6 percentage points less likely than the childless adult to enter homelessness. The Journeys Home sample and model estimates also uncover patterns in the data suggesting that age and country of birth are not statistically important as far as entry into homelessness are concerned. There is also evidence indicating that those with relatively low levels (years) of schooling are more likely to slip out of formal housing circumstances. Somewhat surprisingly, we found that the absence of parenting does not significantly impact on pathways into homelessness. However, those who had been in state care as children are 2 percentage points more likely to enter homelessness, despite a static model finding which suggests that they are no more likely to be homeless overall. Homelessness status is the product of entries into and exits from homelessness, so these apparently puzzling results can be reconciled since exits from homelessness prove to be insensitive to state care status. As expected, risky behaviour (drinking, smoking, and drug use) raises the chances of entering homelessness. However, the effects of ill health on entries into homelessness are mixed. While a long-term health condition increases an individual's likelihood of entering homelessness, having a diagnosed bipolar or schizophrenia condition decreases the probability of slipping out of secure housing and into homelessness. Although this finding is somewhat surprising, given people with mental illness are thought to be especially prone to homelessness, we once again think that it reflects the delivery of treatment and care (even institutionalised care), thereby lowering the chances of entering homelessness compared to those undiagnosed, who could also have other risk factors. Both static and entry models suggest that social support is important in reducing the risk of homelessness. But those housed in any wave are more likely to become homeless in the next six months if they have had a past experience of homelessness. Consistent with the results examining homeless status at a point-in-time (from the static model), median market rents are positively related to entry into homelessness. An increase in the median market rent of $100, which is a 30 per cent increase at the national median weekly rent, lifts the risk of entry by 1.6 percentage points, or from a sample mean of 8 per cent to 9.6 per cent (a 20% increase in risk)