Gun- and Non-Gun Suicides over the Business Cycle

Monday, June 11, 2018: 5:30 PM
Starvine 1 - South Wing (Emory Conference Center Hotel)

Author(s): Jeffrey DeSimone; Tuan Nguyen

Discussant: Susan Averett

Suicide is the one category of mortality for which consensus exists that variation is countercyclical. However, no previous study on the topic has distinguished between suicides committed by gun and other means. We examine how suicide rates, overall and by gun use, vary with unemployment rates (UR) in a 40-year panel of U.S. states. Our results establish that the business cycle fluctuation in suicides stems primarily from those not involving a gun.

In two-way fixed effects population-weighted log regressions that also control for real per capita income, the expected positive relationship between suicide and UR does not emerge until state-specific linear trends are included. In this relatively sparse model, with standard errors clustered by state, a UR increase of one point (roughly the within-state standard deviation) is predicted to significantly raise the overall suicide rate by 1.4%. However, this estimate conceals a significant difference between effects on suicides by gun (0.6%, insignificant at 5%) and other means (2.3%, significant at 1%).

These estimates do not depend on population weights, and are highly robust to imposing various further constraints, including the simultaneous addition of demographic controls, gun-specific homicide rates, and most critically year-by-census division effects. Adding nonlinear state-specific trends reduces the magnitude of the non-gun (and thus overall) effect, to 1.4% (and 1.0%), thereby eliminating the significance of the gun vs. non-gun difference, but does not otherwise alter conclusions. Specifically, estimates remain significant for non-gun and overall suicides, but insignificant for gun suicides, including up to 5th-order polynomial trend terms, matching the nonlinearity in suicide rates over the period, and coefficient magnitudes do not change upon adding terms beyond a quadratic. The invariance of these estimates to omitting time-varying controls other than income is consistent with the implicit assumption that within-state UR variation, specifically relative to other states within the same division and allowing for baseline nonlinear trends, is quasi-random with respect to suicide rates.

Our contributions are twofold. First, we provide an updated estimate of the relationship between business cycle fluctuations and suicide rates. Leveraging identifying variation based on a long, current panel and more defensibly quasi-random than previously used, we corroborate evidence that the UR has a significant positive impact on suicide. Compared to recent evidence, we find the effect to be similar in size, but larger, rather than smaller, during and after the recent economic crisis. Second, we are the first to focus attention in this context on gun suicides, from which more people in the U.S. die on an average day than have been killed in any single mass shooting. Our results indicate that the cyclicality of overall suicides masks the relative insensitivity of gun suicides to changes in economic conditions. This suggests that non-gun suicides will be particularly responsive to policies that effectively mitigate the depth or labor market consequences of economic downturns, and the possibility of deeper underlying differences between the nature of gun- and non-gun suicides.