Extended Data Fig. 3: Placebo test.
From: Increase in suicide following an initial decline during the COVID-19 pandemic in Japan
Panel a represents the results of the placebo test. Using samples from 2016 to 2019, we first randomly allocate treatment dummies within the city across the years. We then regress the suicide rate on the placebo treatment 1,000 times. The mean effect using the placebo sample is around zero, and the estimate of the first outbreak is much smaller than 95% CI of the placebo results and that of second outbreak is much larger than 95% CI of the placebo results. Panel b repeats event study regression using the placebo samples analogously. The estimated coefficient is close to zero in all periods. All regressions include city-by-year fixed effects and city-by-month fixed effects and are weighted by population, and standard errors are clustered at the city level. N = 61,209 (b). Each regression for 1,000 times has different number of observations (a). The separated observations are dropped (see Methods).
