Unemployment and underemployment are causes of suicide

Abstract

Epidemiological studies indicate that labor underutilization and suicide are associated, yet it remains unclear whether this association is causal. We applied convergent cross mapping to test for causal effects of unemployment and underemployment on suicidal behavior, using monthly data on labor underutilization and suicide rates in Australia for the period 2004-2016. Our analyses provide evidence that rates of unemployment and underemployment were significant drivers of suicide mortality in Australia over the 13-year study period. Predictive modeling indicates that 9.5% of the ~32,000 suicides reported between 2004 and 2016 resulted directly from labor underutilization, including 1575 suicides attributable to unemployment and 1496 suicides attributable to underemployment. We conclude that economic policies prioritizing full employment should be considered integral to any comprehensive national suicide prevention strategy.

Fig. 1.. Suicide mortality and labor underutilization rate time series for Australia used to test for causal effects of unemployment and underemployment on suicidal behavior.

See Materials and Methods for details. Gray shading indicates even years.

Fig. 2.. CCM results (embedding dimension, E, 5).

Left panels: Correlations, measured using the Pearson product-moment correlation coefficient, ρ, between the observed unemployment and underemployment rates and unemployment and underemployment rates estimated via cross mapping using the suicide mortality data. Mean correlations (solid lines) and 95 and 50% intervals (light and dark shading, respectively) for each value of L were derived from 10³ libraries randomly sampled (with replacement) from the suicide mortality rate data. Right panels: Null distributions for ρ (at the maximum value of L) derived from 10³ surrogate unemployment and underemployment time series, each generated by randomizing the phases of a Fourier transform of the observed time series (11). The proportions of surrogate time series for which ρ exceeds the observed value (i.e., for the empirical time series, indicated by the vertical red lines) are in the top left corner of each plot. Correlations between the suicide time series and the unemployment and underemployment time series (0.458 and 0.502, respectively) are indicated by the dashed vertical lines (plotted in black).

Fig. 3.. Predictive modeling results.

(A) Empirical numbers of suicides reported per month (red line) and corresponding numbers of suicides predicted from the fitted model (see Results and Discussion). Median predicted numbers of suicides per month (dark gray line) and 95 and 50% intervals (light and dark gray shading, respectively) were calculated from 8000 simulated time series, each generated using a different parameter vector sampled in the Markov chain Monte Carlo (MCMC) analysis (see Materials and Methods). (B) Numbers of suicides per month directly attributable to labor underutilization, estimated by subtracting predicted numbers of suicides per month under a counterfactual scenario in which the unemployment and underemployment rates are equal to zero for all time points from the numbers of suicides predicted assuming the observed unemployment and underemployment rates. Median estimates (dark gray line) and 95 and 50% intervals (light and dark gray shading, respectively) for each time point were derived from expected numbers of suicides calculated for each of the 8000 parameter vectors sampled in the MCMC analysis (see Materials and Methods).