Study cohort: members of share households

This study focuses on people living in share houses in Victoria, Australia. We define share houses as households occupied by two or more unrelated adults who are not in a romantic relationship. In Australia, share housing usually takes the form of individual arrangements between a land lord and a group of tenants; occupants may know each other before moving in together or may begin and remain as relative strangers living in informal arrangements. People living in shared housing are a group characterised by high levels of precarity. They are more likely to be young, casually employed, living in informal arrangements and at risk of homelessness than the broader population.18 In Victoria, the median share household spends 23% of gross household income on housing costs, compared with 14% across all household types;19 17% of temporary visa-holders in Australia were living in a share household at the 2016 census, compared with 4% of the broader population.20 Share housing is often considered as either a transitional housing form on the way to adulthood or a ‘coping mechanism’ for vulnerable households when other forms of family or state support are unavailable21 or where occupants are unable to provide income and rental history documents.22 Recent evidence suggests that these groups are more likely to have lost their jobs or had hours reduced, more likely to be reliant on social welfare payments and more likely to have been born overseas than the general population.1

Study context

This study was conducted in Victoria, the second most populous State in Australia. The study surveys, occurring in June and October of 2020, coincided with a time of considerable disruption to social, economic and health systems in Victoria. While the first positive case of the novel coronavirus was identified in Australia on 25 January 2020, the large-scale impacts of the pandemic were not substantially felt until mid-March 2020. On 30 March, the Australian Government introduced the ‘Job Keeper Payment’ that aimed to help employers keep their staff on pay roll and the ‘Job Seeker Payment’ that served as an emergency CoronaVirus Supplement to existing social welfare payments, immediately doubling the income of many unemployed people.23 The State of Victoria declared a State of Disaster on 2 August 2020, resulting in a night-time curfew, a 5 km limit on distances residents could travel from their homes, restrictions of gatherings in public and private spaces, office and school closures and limitations on allowable time outside the house.24 These restrictions, occurring in the context of a global health pandemic and large-scale economic crisis, present a case study in the impact of simultaneous imposition of housing and employment stress. Furthermore, the substantial government intervention in support payments offers the lens of a ‘natural experiment’ to examine their benefits in mediating the impacts of ‘double precarity’ and exposure to COVID-19.

Conceptual framework

Our conceptual framework is based on the following explanatory variables and mediator variables.

Mediator variables: resources of social support, government support and housing adequacy

Precarity in housing and employment triggered by pandemic containment measures is known to negatively impact mental health. Importantly, several factors mediate the impact of precariousness on mental health, including social support, government support and access to adequate housing.

• Social support: defined as the presence of social ties and frequency of access to emotional and pragmatic support. Social support is often associated with improved mental and physical health, especially as a resource that buffers the harmful impacts of stress exposure.26 27 For example, social support has been shown to mitigate financial hardship via monetary transfers and interpersonal loans in some cases.28

• Government support: while research has documented the ability of government-provided payments to build the resilience of poor and vulnerable households to economic shocks,29 the connection to mental health outcomes is more tenuous. Previous research suggests that social welfare payments need to provide sufficient economic provisions while also alleviating the stigma and psychological impacts associated with receiving benefits to have a protective effect on mental health.30

• Housing adequacy: access to sufficient space and autonomy in a home is important for mental health as non-functioning or inadequate housing is associated with depressive mood.31 Overcrowding in homes can lead to cognitive overload from excess sensory stimuli, a lack of opportunities for retreat and feelings of being surveilled.32 Similarly, previous research has found associations between overcrowding and depression, withdrawal, aggression and psychological distress.33 Living in share housing has been associated with depressive disorders and anxiety, especially for unemployed people.34

Drawing on this literature from health, housing and economics, we hypothesise several channels through which precarity and access to mediating resources impact on each other and on mental health, as illustrated in figure 1.

Survey design and data collection

Data were collected through two waves of an online longitudinal survey, yielding 1052 valid responses in June and 293 valid responses in October. The purpose of a longitudinal design was to test changes in mental health over time, in a period characterised by high levels of COVID-19 shock for many. The survey was open to anyone who had lived in a share household in Victoria at any point between June 2020 and October 2020 and screening questions were used to exclude those who did not meet these criteria. The first page of the online survey contained an informed consent statement and participants acknowledged consent by clicking ‘start’ on the survey. The survey instrument was designed using a variety of standardised demographic, housing, health and financial resilience questions derived from large Australian surveys such as the Household, Income, Labour Dynamics in Australia survey and the Australian Bureau of Statistics Census. See online appendix table A1 for the full survey.

Respondent recruitment occurred through multiple channels. For wave 1 of the surveys, most responses (n=670) were derived from an online survey panel service that targeted a representative selection of share housing respondents currently living in Victoria. The remaining responses (n=382) were targeted through targeted Facebook and Instagram advertisements, Twitter and Facebook messages posted by the University of Melbourne, Tenants Union of Victoria and Victorian Legal Aid and posts on Facebook groups aimed at international students and share houses across Victoria. Wave 2 resurveyed the original respondents, either via an anonymised process managed by the online panel company or through follow-up emails to wave 1 participants automated using the survey programme Qualtrics. In the empirical analysis, we only keep responses for people who appear in both survey waves, which allows us to have a panel of 293 individuals in two time periods. Respondents of the survey are not directly representative of occupants of share houses across all metrics. Compared with share household occupants across Australia, respondents are less likely to report year 12 as their highest level of education than (16% vs 39%), are more likely to be female (55% vs 45%) but are of similar age (median age 35). Despite the relatively high attrition rate, a comparison of means between the included and the excluded participants did not point to any statistically significant difference between the two waves, based on most observable characteristics.

Unlike cross-sectional surveys, this panel survey set-up enables the analysis of individual-level dynamics that are not biased by self-selection, by observing the same individuals repeatedly at two different times of the COVID-19 pandemic. Self-selection would arise due to potential unobservable confounders correlated with both outcomes (mental health) and explanatory variables (double precarity), such as lower motivation or worse work-performance.

Patient and public involvement

Neither patients nor the public were involved in the design, or conduct, or reporting or dissemination of our research. Survey respondents were emailed a copy of research findings if they indicated a desire to receive findings when completing the survey.

Empirical strategy

To investigate the relationship between vulnerability, exposure to COVID-19 and mental health, we proceed in two steps.

We start by analysing the relationship between precarity and respondents’ socioeconomic characteristics using the following regression setup:

(1)

We define double precarity (Precarity_i) as a two-dimensional index reflecting its interaction between employment and housing dimensions for respondent i. This is computed as the summary of the probability of the following conditions, and ranges between 0 and 2:

1. Probability of housing precarity: defined as living in unaffordable housing (paying more than 30% of income on housing costs) and/or renting with a lease of 6 months or shorter.

2. Probability of employment precarity: defined as being casually employed or unemployed.

X_i includes a vector of individual characteristics including gender, age, being low income (weekly income lower than AUD$650), having low education (having completed year 12 or below), being a migrant (namely a temporary visa holder or refugee); ∂_c represents country of birth fixed effects and θ_s denotes sector of employment fixed effects. Given that Precarity_i is an ordinal variable, we estimate this regression’s coefficients using an ordered logit model.

Next, we investigate the effect of Precarity_i, combined with exposure to COVID-19 shocks, to calculate whether either or both affected mental health. Exposure to COVID-19 shocks (COVID-19 shock_i) is measured by a respondent i’s reported impact of COVID-19 in the form of: (i) changing housing conditions (people moving in or out); (ii) decreased earnings; (iii) financial hardship (inability to cover housing and other living costs). We code COVID-19 shock_i as a dummy variable equal to one if an individual had experienced at least one of the above-mentioned shocks. Therefore, the coefficient of COVID-19 shock_i captures the effect of being exposed to COVID-19 shocks (relative to not having experienced any).

We investigate the relationship between mental health, precarity and COVID-19 shocks, using the following regression:

(2)

The dependent variable Mental health worse_i is computed as the summary of the probability of a worsening in mental health in wave 1 and/or wave 2 of the survey. Our measure of mental health was self-reported. Respondents were asked in both waves of the survey ‘Since COVID-19 isolation rules were introduced, would you say that your mental health became: much better, better, did not change, worse, much worse’. We generated a dummy variable equal to 1 if respondents answered worse or much worse in each wave. Given that Mental health worse_i is an ordinal variable, we estimate this regression’s coefficients using an ordered logit model. Regressions (1) and (2) cannot establish a causal relationship between outcomes and explanatory variables, and should be interpreted as correlations.

Finally, to provide insights on plausible mediating factors that may mediate the negative relationship between precarity and health outcomes, we assess the role played by:

1. Sufficient government support (Gov support_i) targeted to mitigate negative COVID-19 effects, measured as respondents’ self-evaluated sufficiency of support derived from accessing the packages offered by the government to assist financially those affected by COVID-19. Specifically, we assign a dummy variable equal to 1 for each respondent answering ‘somewhat agree’ or ‘strongly agree’ to the question ‘The resources I have accessed in response to COVID-19 are sufficient to make a substantial difference to my financial security over the next 3 months’.

2. Social support (Social support_i), measured as the presence of community or family networks used as risk-coping mechanisms and their frequency of access during the pandemic, modified from.35

3. Housing inadequacy (Housing inadequacy_i), computed as a multidimensional index drawing on perceptions of privacy, use of space and overcrowding, modified from Campagna.36

We undertake a mediation analysis to examine the extent to which the association between precarity and a worsening in mental health occurs directly, and the extent to which it occurs through housing inadequacy and lack of social support.

Following VanderWeele,37 we utilise the following regression setup:

(3)

where M represents either Social support_i, Gov support_i, Housing inadequacy_i or COVID-19 shock_i, and X_i is a vector including the above-mentioned set of controls as well as an indicator of exposure to COVID-19 shocks (when M is not COVID-19 shock_i).

This approach allows us to compute: (i) the natural direct effect (NDE), capturing how much precarity would affect mental health if we were to disable the relationship between precarity and the mediators; (ii) the natural indirect effect (NIE), which can be conceived as the effect on mental health of the mediator, keeping Precarity_i fixed; (iii) the total effect (TE) representing the summary of NIE and NDE, which can be defined as how much mental health would change overall for a change in precarity, accounting for the mediators’ effect.