Do downturns cause desperation? The effect of economic conditions on suicide rates in Canada

Abstract

There is conflicting evidence concerning the impact of macroeconomic conditions on suicide rates. To help resolve this pertinent question, we present evidence using Canadian data. We estimate feasible generalized least squares models of annual gender-specific suicide rates in the working age population (aged 25–64) using data from each of the 10 Canadian provinces over the period 1982 to 2007. We allow for heteroscedasticity across provinces and first-order autocorrelation common to all provinces. We posit that suicide rates in this population are a function of macroeconomic conditions (current and lagged unemployment rates and real per capita GDP) and other determinants that might be correlated with macro conditions, such as physician supply. We find that different factors affect suicide rates across genders and that some of the results are sensitive to the specification of the model we use and the regressors included. Generally, economic conditions affect men more than women; suicide rates are counter-cyclical and a higher supply of psychiatrists in a province is correlated with lower suicide rates.

Keywords:

• suicide
• business cycle
• macroeconomic conditions
• physician supply

JEL Classification:

• I18
• I19

Acknowledgements

We thank Martin Dooley, Lonnie Magee, Brian Ferguson, Catherine Deri Armstrong, Jeffrey Smith, Thomas Parker, an anonymous referee and seminar participants at the Department of Economics of the University of Waterloo, Canadian Economic Association meetings and Canadian Health Economics Study Group for helpful comments and suggestions on an earlier version of the article. We also thank Minsup Shim for invaluable research assistance. All remaining errors are ours.

Notes

¹ Additionally, Lester and Yang (1997) provide an excellent survey of the literature on suicide up to the late 1990s.

² We experimented with various distributed lags and included lags up to 10 years. There were no statistical differences in the models and estimate precision was enhanced by limiting our lags to one period.

³ The Hausman test needs to be modified due to the presence of heteroscedasticity and autocorrelation. Wooldridge (2002, p. 291) describes how to modify the test.

⁴ We modify the RESET test due to the presence of time trends and province-level fixed effects. We regress the de-trended variables on each other and obtain the square, cube and fourth power of the predicted value of the dependent variable from the regression model. We include these three variables in the regression of the de-trended variables. The p-values reported in the table come from joint hypothesis tests on the three variables obtained from the predicted value. The logic from this test follows Wooldridge (2002, Section 11.2.2).

⁵ We also ran Dickey–Fuller tests on each of the dependent variables to check if the reason the panel-specific tests rejected the hypothesis of a unit root was because of a single province. This is not the case. Using the Dickey–Fuller tests, we can most often reject the hypothesis of a unit root at the 1% significance level, and always at the 10% or better. There are seven exceptions: Women’s suicide rate in British Columbia, Saskatchewan and Ontario and Men’s suicide rate in British Columbia, Saskatchewan, Ontario and Newfoundland.