EA & AI Safety Japan community member
Le Wagon Tokyo—data science & AI (Japan)
MA in Psycholinguistics from HHU Dussedorf (Germany) & Chiba University (Japan)
BA from Kyiv Taras Shevchenko University (Ukraine)
Yulia Chekhovska
Karma: 90
Thank you! I will correct it.
Sorry for this, I will change it, but based on everyone’s comments, I think I can make many more changes in the article or write part 2 later, make it about this year’s 2025 results.
( We haven’t published them openly yet and Japan police hasn’t updated their data either, so I could only analyze the 2024 intervention.)
Thank you so much!
The hotline is indeed considered the strongest channel at the moment at my nonprofit for suicide prevention impact, since people who call it do so of their own will and usually in a state of emergency. I would argue that the school workshops are very close to being effective since they also give an opportunity for a 1-on-1 conversation with a lifeline volunteer, albeit face-to-face. But I don’t know how to attribute the percentages here. The GiveWell cost-effectiveness article mentions moral weights, but they are more like philosophical moral judgments of questions about the person affected.
https://www.givewell.org/how-we-work/our-criteria/cost-effectiveness/moral-weights
In data science for example the transformers are analyzing a sentence token by predicting the next token in line based on the weights, so if I follow the logical I need to assign cost effectiveness weights to decide how much each channel contributed to effectiveness. I wish there were more studies on this online.
Thank you for your feedback. I agree with you that separating each channel and evaluating its cost effectiveness is better if my nonprofit decides to evaluate the program and decide on what to scale up or down. About overestimating the indirect outreach—very very true. But I needed to start with at least a number that is closer to reality and based on logical assumptions and the human indifference factor wasn’t helping the case (school simply not replying to our emails despite having received the packages or just sending a simple reply with no follow up).
As I also wrote above, there is a risk of duplicates, we might distribute materials and also be invited to a school to deliver a suicide prevention workshop (a training of sorts). In that case, the numbers would look much lower if we counted both channels separately, because the same people are being reached.
@Cat🔸 below wrote about analyzing youth suicide distribution across the regions and countering it with the school outreach in regions, which seems like also a more topical approach for the future analysis.
Thank you! I agree with @NickLaing , for a wealthy country with GDP per capita of $50,000, this would mean being willing to spend the entire national income to extend everyone’s healthy life by one year, which does seem generous. In practice GiveWell’s bar is much higher like around $100-200 per DALY (roughly 170-340× stricter than WHO’s threshold for wealthy countries).
I got the insights about Givewell cost effectiveness in suicide prevention here but I’m pretty sure it is not comparable to my intervention calculations, pesticides ban being a much clearer intervention to follow than workshops and suicide prevention hotline.
https://www.givewell.org/research/incubation-grants/CPSP-general-support-jan-2021
That’s a good idea, as it will give us even more insight into which regions we should target more in order to increase impact. The regional data also exists on the Japan Police Agency website, so this should be feasible. Thanks!
As for the schools news about a suicide—it’s hard to coordinate to receive this news for mostly human-related reasons (most point of contacts usually only contact my nonprofit if they need a workshop, they never give us any feedback).
Thank you! I had to proceed with so much caution as everything evolving around mental health data collection is a huge taboo in my nonprofit. But it gives me hope that I can produce a better report on cost effectiveness of the 2025 program already in 2026.