The impact of social housing on mental health: longitudinal analyses using marginal structural models and machine learning-generated weights

Int J Epidemiol. 2018 Oct 1;47(5):1414-1422. doi: 10.1093/ije/dyy116.

Abstract

Background: Social housing may provide an affordable and secure residential environment, but has also been associated with stigma, poor housing conditions and locational disadvantage. We examined the cumulative effect of additional years, and tenure security (number of transitions in/out), of social housing on mental health in a large cohort of lower-income Australians.

Methods: We analysed a longitudinal panel survey that annually collected information on tenure and health from 2001 to 2013. To address the time-varying effect of previous health on social housing occupancy, we used marginal structural models. Stabilized inverse probabilities of treatment weights were generated using ensemble learning to improve prediction. To address remaining residual imbalance across covariates, double adjustment was made by additionally including baseline covariates in models. Mental health was measured using the Mental Health Short-Form summary measure of the SF-36 (MH), and psychological distress was measured using the Kessler Psychological Distress Scale (K10).

Results: People who had continuous exposure to social housing had worse mental health on average than people continuously occupying other tenures. The worst mental health outcomes, however, were observed for people who made multiple transitions. Mental health deteriorated and psychological distress increased with number of transitions: MH -1.04 [95% confidence interval (CI) -2.16; 0.09) and K10 0.56 (95% CI 0.12; 1.00). Estimates are in the order of 6% (MH) and 9% (K10) of one standard deviation for each measure.

Conclusions: The more transitions people made in/out of social housing, the greater the impact on mental health and psychological distress, supporting the case for provision of more stable forms of social housing.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adult
  • Aged
  • Australia / epidemiology
  • Female
  • Humans
  • Longitudinal Studies
  • Machine Learning*
  • Male
  • Mental Health*
  • Middle Aged
  • Models, Statistical*
  • Poverty
  • Public Housing / statistics & numerical data*
  • Social Stigma*
  • Stress, Psychological / epidemiology*