Download all data pages
PDF 5.6Mb
Data

Regression risk models for selected census variables

Generally, results from the modelling show important differences in the relationship between deaths by suicide and the different socioeconomic factors, relative to comparison groups, as seen in the forest plot below.

Univariate and multivariate competing risk models were used (Fine and Gray, 1999). Results of sex stratified models are also reported, these are multivariate models split by males and females to investigate the interactions within the sex. 

See Technical notes for further information on the data and methods used.

Estimates presented are hazard ratios for the group of interest compared with a reference group. Reference group values are indicated as the dotted line at 1. A hazard ratio (HR) indicates how many times higher the probability of an event is in one group of people with a particular characteristic than in another group without that characteristic, after adjusting for other factors in the model. The size of the reported hazard ratio indicates the strength of the relationship a social factor has to deaths by suicide, relative to the reference group.

Ninety-five per cent (95%) confidence intervals are also presented to indicate the statistical precision and significance. The result is interpreted as having a statistically significant impact (that is, not due to chance) if the interval does not cross the value of 1.

This chart shows the output from competing-risks regression models to explore the association between socioeconomic factors and deaths by suicide. For simplicity and ease of understanding, the model estimates are reported as hazard ratios.

Results from four models: univariate, multivariate, stratified: males and stratified: females can be displayed in this chart. The univariate model does not adjust for the other socioeconomic factors, while multivariate model adjusts for all other factors. The stratified: males and stratified: females models are multivariate models for only males and females, respectively.

Which social factors are associated with an increased risk of death by suicide?

The modelling carried out includes only a subset of known factors that may influence deaths by suicide. Results from this analysis need to be interpreted with caution and within the context of the information provided. For example, due to data quality and availability, associated factors such as 'mental health status', 'acute or chronic substance use', and 'past-history of self-harm' are not included in this modelling.

Results of the multivariate analysis showed that from September 2011 to December 2017, when adjusting for other factors in the model:

  • Males had a higher risk of death by suicide than females (hazard ratio of suicide among males was 3.1 times higher than females).
  • Those who were widowed, divorced or separated (HR = 1.95; 95% CI 1.79 to 2.12), and those who never married (HR = 1.82; 95% CI 1.69 to 1.96), as well as those in lone person households (HR = 1.72; 95% CI 1.57 to 1.87), had a higher risk of suicide compared with people who were married or in a de facto relationship and couples with no children, respectively
  • The hazard rate of suicide among people who were unemployed or not in labour force were both 1.8 times higher compared with those who were employed.
  • The univariate analysis estimated that Indigenous Australians had approximately double the risk of death by suicide when compared to non-Indigenous Australians (HR = 2.05; 95% CI 1.81 to 2.31).  This is consistent with the 2.06 ratio of age-standardised death rates between Indigenous and non-Indigenous Australians reported by the ABS Causes of death Table 11.4.  However, when controlling for the sociodemographic factors included in the multivariate analysis the relative risk of death by suicide for Indigenous Australians was estimated to be 1.3 times the rate for non-Indigenous Australians (HR = 1.31; 95% CI 1.13 to 1.51).  The 74% reduction in the difference in suicide risk for Indigenous Australians between the multivariate and the univariate model demonstrates the significant effect of the sociodemographic factors for suicide risk that are included in the multivariate model on Indigenous Australians.
  • The hazard rate of suicide among people whose highest educational attainment was secondary education or lower (including no education) or diploma and certificate were both 1.3 times higher than those who had at least a bachelor degree.
  • Those with lower incomes had higher risk of suicide, with HRs of 1.59 (95% CI:1.45 to 1.74), 1.70 (95% CI: 1.57 to 1.85) and 1.28 (95% CI: 1,19 to 1.38) for low, medium-low and medium-high income earners compared with high income earners, respectively.
  • People who worked as machinery operators and drivers (HR = 1.51; 95% CI 1.36 to 1.69) or labourers (HR = 1.36; 95% CI 1.22 to 1.52) were more likely to die by suicide compared with managers and professionals those aged 35–44 (HR = 1.95; 95% CI 1.79 to 2.12) and 45–54 (HR = 1.82; 95% CI 1.69 to 1.96), had a higher risk of deaths by suicide compared with the reference age group (25–34 years); however, there was no difference for those aged 55-64.

When separated by sex and adjusting for other factors, important differences were:

  • Females who needed help with core activities of daily living had 2.4 times higher hazard rate of suicide compared with females who did not need help with core activities of daily living (Stratified: female model). There was no significant difference observed for males.
  • Males who were single parents had 1.3 times higher hazard rate of suicide compared with males in a couple without children (Stratified: male model); however, there was no difference between females with the same household compositions.

This analysis was carried out in consultation with the Australian National University, the University of Melbourne and the University of Western Sydney.

Austin PC & Fine JP 2017. Practical recommendations for reporting Fine‐Gray model analyses for competing risk data. Statistics in Medicine 36: 4391–4400.

Australian Bureau of Statistics (ABS) 2021. Table 4. Estimated Resident Population, States and Territories (Number) [time series spreadsheet], National, state and territory population, accessed 24 August 2021.

ABS 2019. Microdata: Multi-Agency Data Integration Project, Australia, ABS website, accessed 17 August 2021.

ABS 2016. Research Paper: Death Registrations to Census Linkage Project - A Linked Dataset for Analysis, Mar 2016, ABS website, accessed 17 August 2021.

Biddle N & Marasinghe D 2021. Using census, social security and tax data from the Multi-Agency Data Integration Project (MADIP) to impute the complete Australian income distribution. Tax and Transfer Policy Institute – Working Paper 8/2021.

Fine J & Gray R 1999. A proportional hazards model for the subdistribution of a competing risk. Journal of the American Statistical Association 94(446): 496-509.

Hernan MA 2010. The hazards of hazard ratios. Epidemiology 21(1): 13–15.