Background Poor housing affordability affects around 10% of the Australian population and is increasingly prevalent. The authors tested two hypotheses: that cumulative exposure to housing affordability stress (HAS) is associated with poorer mental health and that effects vary by gender.
Methods The authors estimated the relationship between cumulative exposure to HAS and mental health among 15 478 participants in an Australian longitudinal survey between 2001 and 2009. Individuals were classified as being in HAS if household income was in the lowest 40% of the national distribution and housing costs exceeded 30% of income. Exposure to HAS ranged from 1 to 8 annual waves. Mental health was measured using the Short Form 36 Mental Component Summary (MCS) score. To test the extent to which any observed associations were explained by compositional factors, random- and fixed-effects models were estimated.
Results In the random-effects models, mental health scores decreased with increasing cumulative exposure to HAS (up until 4+ years). This relationship differed by gender, with a stronger dose-response observed among men. The mean MCS score of men experiencing four to eight waves of housing stress was 2.02 points lower than men not in HAS (95% CI −3.89 to −0.16). In the fixed-effects models, there was no evidence of a cumulative effect of HAS on mental health; however, lower MCS was observed after a single year in HAS (β=−0.70, 95% CI −1.02 to −0.37).
Conclusions While average mental health was lower for individuals with longer exposure to HAS, the mental health effect appears to be due to compositional factors. Furthermore, men and women appear to experience cumulative HAS differently.
- Housing affordability
- mental health
- longitudinal studies
Statistics from Altmetric.com
Housing and health are inextricably linked in a relationship that spans the lives of individuals. Housing has been shown to influence both physical and mental health via direct or indirect pathways. Direct pathways have, for example, been shown between warm, dry houses,1 ,2 overcrowding3 or the presence of toxins4 and both physical and mental health. Housing and health have also been shown to be related through more indirect pathways, where a housing characteristic may influence other outcomes which in turn influence health outcomes. Such diffuse relationships have been shown for housing tenure,5 ,6 neighbourhood and place,7 incidence of local crime8 ,9 or as is the focus of this paper—housing affordability problems.10–12
In the current era of housing affordability ‘crisis’ that has been documented across many post-industrial countries including the USA13 and Australia,14 the health effects of poor housing affordability are of special relevance and a critical problem for governments. Using the example of Australia, housing affordability stress (HAS) is estimated to affect >850 000 households15 and, based on the Australian average household size estimate of 2.5 persons,16 this equates to around 2 150 000 individuals (∼10% of the population). Not only are increasing numbers of individuals experiencing HAS in Australia but it is also likely that the magnitude of, and time spent in, unaffordable housing will also increase. Many individuals experiencing poor housing affordability are also vulnerable to the effects of continuous time periods of HAS. Importantly, while there is an existing and expanding evidence base on the health effects of poor housing affordability, relatively little is known of the additional health effects of time spent in housing stress.
As is often stated, housing is a key determinant of health.4 Housing affordability affects health through both the housing that can be obtained as well as the income remaining for other household expenditure, for example, food, transport, medical care. For many households, housing costs are a ‘foundation’ expenditure prioritised before other non-shelter expenditures, therefore one important pathway being established between housing affordability and health relates to the influence of the pressures of meeting housing costs upon mental health. Nettleton and Burrows17 found a clear association between mental health and mortgage affordability, and following this, Taylor et al18 identified measurable effects of ‘unsustainable housing commitments’ on the mental health of household heads. Taylor et al's18 work also revealed a gender difference in mental health effects for men and women. They found that developing housing affordability problems was more likely to be detrimental for men, whereas for women, peak detrimental effects were likely to be experienced once affordability problems became longer term. Similarly, Pevalin et al11 observed women's mental health was more likely to be affected by housing improvements but not deterioration. These findings point to a gender difference in the effects of housing affordability problems and to a possible difference in the strategies that men and women use to deal with them. It has been suggested that gender differences in mental health arise from different help-seeking behaviours, biological differences and the different ways in which women and men acknowledge and deal with distress.19 Men and women may also respond to stressors differently.20 Wells and Harris21 in their longitudinal examination of housing and mental health found that social withdrawal mediates between housing conditions and mental health, whereby “social withdrawal influences psychological distress and that social withdrawal is influenced by housing quality” (p 74). Nettleton and Burrows17 also suggested that men experiencing mortgage arrears and repossession are more likely to make more GP visits than women.
Building upon earlier work11 ,18 ,22 that established associational and causal links between housing affordability and mental health, this paper investigates the effect of cumulative housing stress among the Australian population. We examine if the effect of continuous periods of HAS follows a dose-response relationship similar to that described by Marsh et al,23 where sustained exposure to housing deprivation increased the probability of ill health. In addition, we test if there is a gender difference in mental health effect of HAS over time.
We test two hypotheses:
That there is a dose-response relationship for HAS and mental health, such that more time in HAS is increasingly worse for mental health and
that the dose-dependent relationship between HAS and mental health for men and women is different.
The Household, Income and Labour Dynamics in Australia longitudinal study follows Australian households and individuals across annual survey waves. The data set is based upon a nation-wide probability sample of individuals aged 15+ years, focused on income, employment, health and well-being, with data collected using face-to-face interviews and self-completion questionnaires, as described by Watson.24
The analysis described in this paper is based upon the 75 865 responses of 15 478 Australian Household, Income and Labour Dynamics in Australia participants between 2001 and 2009, for whom housing status and mental health score, and information on each covariate, were available.
The Mental Component Summary (MCS) of the Short Form 36 measure was used as the outcome measure of analysis. The Short Form 36 is one of the most widely used self-completion measures of health status25 across a number of dimensions, such as Physical Functioning, General Health and Mental Health. A low MCS score indicates frequent psychological distress, social and role disability due to emotional problems, and poor self-rated health.
HAS was defined in this analysis as occurring when a household was positioned in the lowest 40% of the equivalised disposable income distribution and paying >30% of gross household income in rent or mortgage costs.26 ,27 The cumulative measure of HAS was based on the number of consecutive annual waves an individual was recorded as being in HAS. A period of HAS had to have been preceded by at least one wave in which the individual was not in HAS to provide an accurate measure of cumulative exposure. Due to small numbers, people with four, five, six, seven or eight waves of consecutive exposure were combined into a single group. The reference group comprised individuals who were not in HAS for at least two consecutive annual survey waves. The structure of this variable is illustrated in figure 1.
Potential confounders were identified from existing literature and consideration of the likely relationship between exposure to cumulative HAS and change in MCS. All models were adjusted for age (centred on mean). Models that were not stratified by sex were adjusted for it. The analysis was also designed to control for the anticipated influence of unemployment and existing health conditions on likelihood of individuals being in HAS and having poor mental health. Adjustment was made for respondents reporting a long-term health condition (No/Yes) and for the number of survey waves spent unemployed (0–9 years).
Of the 100 783 observations recorded in the data set between waves 1 and 9, 75 865 (75%) were included in these analyses. Observations were excluded where health information was missing or for records that did not have the appropriate pattern of HAS information to be included in either the reference or a cumulative exposure group.
Within the data set, at each wave, 9%–15% of income values were imputed by the data custodians as detailed by Watson.24 This imputation was undertaken to address bias likely to result from missing data. We examined the possible effect of imputation by comparing results generated from both imputed and un-imputed data (results not shown). These tests produced results of similar magnitude and significance. We concluded that it was appropriate to use imputed values in our analytical sample. The results presented in the paper are hence based upon imputed income data.
Longitudinal regression models with both random-effects and fixed-effects estimators were run in Stata V.11.0. The coefficients obtained from the random-effects models reflect a weighted average of between-subject differences (difference in mental health score between two subjects based on their exposure to HAS) and within-subject differences (change in mental health score within a subject as their exposure to HAS changes). The fixed-effects model coefficients reflect only the within-subject differences. We have used both in this paper to compare the extent to which differences between people or, in other words, the composition of people who experience cumulative periods of poor housing affordability, explain any observed associations. Random-effects models describe the relationship between cumulative exposure to HAS and mental health, while fixed-effects models isolate the non-compositional elements, indicating the extent to which it is characteristics of the population that explain any observed associations in the random-effects models.
Rather than assessing how change in the exposure variable predicts change in the outcome, we have defined our exposure variable over time so that we can see the impact on mental health of strictly defined patterns of exposure.
We estimated the association between cumulative exposure to HAS with changes in mental health adjusted for age, years of unemployment and long-term health condition (and sex in the random-effects models). Models were estimated for all people in the sample. We then included an interaction term in the random-effects models to test for effect modification by sex and subsequently stratified the results of both models by sex.
Description of the analytic data set
The mean MCS among the sample population was 49 (SD=10). Across both genders, mean mental health score increased with age. Men and women with a long-term health condition reported lower mental health scores (table 1). Mental health decreased with each additional wave a person was categorised as being in HAS, and time spent unemployed was associated with a lower MCS.
Table 2 shows self-assessed health scores relative to the reference group (no housing stress for two consecutive surveys) for both the random- and fixed-effects models. The random-effects model shows that, compared with the reference group, MCS scores were, on average, lower for individuals exposed to HAS. People who experienced four or more waves of consecutive HAS had a MCS 1.19 points lower than the reference group (95% CI −2.34 to −0.05; random-effects model). The p for trend test provided strong evidence against the null hypothesis of no linear relationship between length of exposure and impact on mental health.
There were noticeable differences in the results between the random- and fixed-effects models. The random-effects results are statistically significant at each of the four levels of cumulative HAS, and the MCS decreased monotonically until the category representing 4+ years, where dose appeared to no longer have an effect. However, using a fixed-effects modelling strategy, the association between HAS and mental health was only significant in the year of entering housing affordability stress and further, the mental health effect in this year was small (0.70). When individuals had been in HAS for longer periods, effect sizes were smaller, and there was little evidence to reject the null hypothesis of no association.
Although the interaction terms between gender and cumulative exposure to housing stress were not statistically significant, stratified models were estimated on the a priori assumptions that men and women were likely to experience HAS in different ways18 and that the interaction test may have been underpowered given the relatively small number of men and women who had experienced long-term HAS (4+ consecutive years) in the analytical sample (39 men and 78 women).
Table 3 shows the gender-stratified results. It highlights a clear difference in the pattern of dose-response for men and women. Considering the random-effects model, we see a (highly significant) dose-response relationship for men, with mental health scores decreasing with each additional year of HAS, from −0.84 at 1 year to −2.02 for those with 4+ years HAS. The results for women are noticeably different, with the largest negative effect for those with 1 year of HAS (−1.28) and a slightly smaller decline in average mental health score for those with 2 years of HAS (−1.17), and though the general pattern for 3 and 4+ years of HAS is for continued improvement to average mental health, these results are not significant. Importantly, the fixed-effects models stratified by gender reveal very different patterns. In this case, we find no significant differences from the reference group for men and for women, only those with 1 year in HAS show a significant reduction in mental health.
These results using both random- and fixed-effects modelling approaches indicate much about the way that cumulative HAS acts on individual mental health. We see that, when considered across a population, the time spent in HAS is an indicator of self-assessed mental health, and this relationship is dose dependent. This result holds even when age, sex, time unemployed and the presence of health conditions are taken into account. Importantly, though these random-effects results compare and describe averages in each level of HAS, they are likely to be confounded by factors that differ between individuals and have not been adjusted for in the regression models, for example, prior housing affordability and mental health. When we use a fixed-effects approach, examining only within-person changes in mental health rather than also the differences between people at different levels of exposure to HAS, this confounding by unaccounted-for differences between people is removed. Here, only a small mental health effect can be attributed to HAS, and only in the year they enter HAS. This indicates that while the immediate effect of entering HAS is likely to have an intrinsic and detrimental effect on mental health (as supported by Bentley et al28), the association between mental health and cumulative exposure to HAS is more directly attributable to the characteristics of individuals exposed to HAS for consecutive years. We therefore suggest that the most appropriate explanation for differences in mental health with increasing cumulative HAS exposure will be compositional, rather than causal. That is, characteristics of people who experience HAS account for the dose-response effect observed in the random-effect models.
Reflecting on the gender-stratified results with this in mind, our analysis indicates that, at the cohort level, cumulative exposure to HAS appears to have a differential impact on men and women. Men who experienced 4+ years of HAS had 2.02 points lower MCS on average than the reference group, whereas the average mental health score of the corresponding group of women was 1.28 points lower. Interestingly though, the dose-response trend established for men was absent or possibly even reversed for women. Importantly, the within-person analysis shows almost no mental health decrease directly attributable to cumulative HAS (other than a highly significant but small negative effect for women in the year they entered HAS). Overall, our results indicate that the effect on mental health of HAS is due to the composition of the population of people who experience cumulative years of poor housing affordability and that this is likely to be different for men and women.
This study has several important strengths and some limitations. This is one of the few studies that attempts to quantify housing's influence over health in relation to gender. Little research has been previously conducted on length of exposure to HAS and the likely differential impact on men's and women's mental health. Our study is the first to explore such trends over an extended period, 9 years. Further we begin to establish evidence for the indirectness of the relationship, and this represents a promising future research direction. An additional strength is our utilisation of a large and robust national longitudinal data set and a causally focused methodology in this multidisciplinary study, benefiting from the methods and concepts of both social epidemiology and housing research.
The research is based upon secondary data, which necessarily contains some limitations. We have not accounted for potential clustering of individual mental health within households. Additionally, the number of people experiencing longer periods of HAS is small, particularly in the gender-stratified models. There is, however, significant scope for future research to expand the analyses as additional waves of data become available, to more definitively identify trends.
Though these findings suggest that pathways between HAS and mental health are different for men and women, this paper highlights the importance of future research to unpack the role of housing affordability in alleviating broader disadvantage. Substantial value would come from in-depth qualitative interviews with men and women who have had long periods of exposure to HAS. Such qualitative work might also make progress towards better understanding of the indirect pathways by which housing affordability influences health outcomes. While we have shown an overall influence of dose, more work is required to unpack the details of the important stress pathways in operation.
The differences observed in this study may appear small in magnitude; however, the size of the measured effects need to be considered in the light of the impacts that even small shifts in population means can have on the numbers of individuals at the marginal ends of those populations. This means that any changes in social determinants that create small shifts in population distributions can potentially increase numbers of people who are at risk.29 In the case of this analysis, increased exposure to HAS across a population is likely to increase the proportion in a population at risk of mental health problems who might require treatment, for example, depression or anxiety. Furthermore, small effects experienced by large populations, such as the >2 million individuals we estimate to be in HAS in Australia, are of great public health and policy importance.
This study contributes to an emerging empirical evidence base, which seeks to understand the relationship between housing and health. The fact that ongoing housing affordability problems are shown to result in significant and measurable decreases in mental health across a population is further evidence of the interrelatedness of housing and health, an interrelatedness that is often poorly reflected in the largely separate housing and health policies in Australia and across many post-industrial nations. Critically, this analysis demonstrates that it is important to understand how social determinants, such as housing affordability, affect men's and women's health differently if we are to implement useful policies for reducing health inequalities.
What is already known on this subject
Housing is a key social determinant of health and the main expenditure for a majority of households in post-industrial countries.
Poor housing affordability affects mental health, via the stress of housing payment problems, and men and women appear to respond to housing affordability stress differently.
What this study adds
This study contributes to understanding of the longer term mental health effects of poor housing affordability for men and women living in Australia by focusing on cumulative exposure.
Using longitudinal data, we find evidence of a negative relationship between poor housing affordability and mental health that is dose dependent and different for men and women; however, this cumulative effect appears to be largely due to composition factors.
There is strong evidence of a negative association between housing affordability stress and mental health in the first year in which an individual lives in unaffordable housing, and this is not explained by confounding by the other characteristics of people who enter housing affordability stress.
We gratefully acknowledge Anne Kavanagh, Andrew Beer, Terry Hartig, Julie Simpson and Laurence Lester and the two peer referees for their valuable comments and advice in the preparation of this paper.
Disclaimer This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and 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 author and should not be attributed to either FaHCSIA or the Melbourne Institute.
Funding The work was supported by the National Health and Medical Research Council Capacity Building Grant (The Australian Health Inequities Program) as well as a Key Centre for Women's Health in Society Director's Start-up Grant and a Flinders University Establishment Grant.
Competing interests None.
Ethics approval The Household, Income and Labour Dynamics in Australia Survey was approved by the Faculty of Business and Economics Human Ethics Advisory Committee at the University of Melbourne.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.