I really like this article. I was going to share the standard EA cost effectiveness graph (the one at the top) and I thought “do we really know this is right?” and then there was this article and I could look through and kick the tires a bit.
So thanks to @Benjamin_Todd and anyone else involved for writing it and thanks to @Toby_Ord and anyone else involved for the research that started the discussion off.
Doing research on the things “everyone knows” seems very valuable.
Should we just be finding 80k pages and reposting them on the forum?
I think the world is better when we are well-versed in key research. Posting these to the forum (with proper attribution) seems to make that more likely.
In some cases that will come with IP concerns, since someone may have their work behind a paywall, but that seems very unlikely to be the case here.
Also 80k doesn’t have a comments section.
However I asked Benjamin afterwards and if he says there is an issue, I will grumpily update.
Executive summary: Studies across multiple domains show that the most effective interventions are typically 10-100 times more cost-effective than average interventions in the same area, but this likely overstates true forward-looking differences.
Key points:
Data from global health, education, climate change, and other areas consistently show heavy-tailed distributions of intervention cost-effectiveness.
The most effective 2.5% of interventions are typically 20-200x more cost-effective than the median and 8-20x more than the mean in backward-looking studies.
Forward-looking differences are likely smaller due to regression to the mean, unavailability of best interventions, and other factors.
The author estimates the true top 2.5% of measurable interventions are 3-10x better than the mean, with non-measurable interventions potentially adding another 2-10x factor.
Using data to identify top interventions may boost impact 3-10x compared to random selection within a cause area.
Cause selection likely matters more for impact than intervention selection within a cause.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, andcontact us if you have feedback.
This has already been posted and discussed on the forum: https://forum.effectivealtruism.org/posts/seFH9jcH3saXHJqin/data-on-how-much-solutions-differ-in-effectiveness
Oooh rats, I looked for that. Thank you.
I really like this article. I was going to share the standard EA cost effectiveness graph (the one at the top) and I thought “do we really know this is right?” and then there was this article and I could look through and kick the tires a bit.
So thanks to @Benjamin_Todd and anyone else involved for writing it and thanks to @Toby_Ord and anyone else involved for the research that started the discussion off.
Doing research on the things “everyone knows” seems very valuable.
Obvious meta question:
Should we just be finding 80k pages and reposting them on the forum?
I think the world is better when we are well-versed in key research. Posting these to the forum (with proper attribution) seems to make that more likely.
In some cases that will come with IP concerns, since someone may have their work behind a paywall, but that seems very unlikely to be the case here.
Also 80k doesn’t have a comments section.
However I asked Benjamin afterwards and if he says there is an issue, I will grumpily update.
Executive summary: Studies across multiple domains show that the most effective interventions are typically 10-100 times more cost-effective than average interventions in the same area, but this likely overstates true forward-looking differences.
Key points:
Data from global health, education, climate change, and other areas consistently show heavy-tailed distributions of intervention cost-effectiveness.
The most effective 2.5% of interventions are typically 20-200x more cost-effective than the median and 8-20x more than the mean in backward-looking studies.
Forward-looking differences are likely smaller due to regression to the mean, unavailability of best interventions, and other factors.
The author estimates the true top 2.5% of measurable interventions are 3-10x better than the mean, with non-measurable interventions potentially adding another 2-10x factor.
Using data to identify top interventions may boost impact 3-10x compared to random selection within a cause area.
Cause selection likely matters more for impact than intervention selection within a cause.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.