Founder of Overcome, an EA-aligned mental health charity
John Salter
Karma: 2,077
Which donations do you think had the highest EV in retrospect?
You and Jeff could’ve easily scaled back your donations when your income decreased—no one would’ve questioned it, and it would’ve made total sense. But instead, you stuck with giving half. The only logical explanation is extreme selflessness! It’s really inspiring to see. Thank you for writing this post :)
Vikram Patel and Shekhar Saxena are pretty prolific researchers in this space too. Both are relatively easy to contact—I reached out to both and both got back to me pretty quickly
Fascinating idea! I suspect general access isn’t going to be cost-effective enough, but there’s likely a niche demographic with a niche issue that this would work wonders for. I’d love to hear ideas anyone has for what those niches might be.
Impact = intervention chosen—what would have happened anyway
EAs often ignore the latter altogether, especially in terms of donation / volunteer source.
Overcome: Growth and Marginal Cost-Effectiveness Data
This is helpful, thank you!
I think this post would be more compelling if you shared more data (e.g. cost per participant, leading indicators of impact), how much money you got from EAIF etc
Thank you for thinking critically about my work! You’re right, it is not a direct comparison.
It shows an effect size of just over 0.6. The typical for most psychotherapies is 0.8. (see the Perplexity.ai summary below of the PTSD meta-analyses in the literature)
I did 0.6 / 0.8, which is 0.75. That equates to 75%.
As this is pretty approximate, especially given that it didn’t directly compare the same groups against one and other. I included the ~ before 75% to show that it shouldn’t be used as a precise figure. In hindsight, I regret not making this more explicit.
That being said, the near equivalency between laypersons and trained therapists is widely accepted. Every single EA mental health charity uses laypersons rather than professional therapists for this reason
and maybe newcomer?
I am! Just under two years delivering psychotherapy interventions, ~5 years in mental health more generally
Could you tell me more about the length, intensity, and duration of a typical treatment program?
We offer a minimum of six weeks, with no arbitrary cap. It’s once (or rarely twice) a week for ~1 hour at a time. I’d suggest that six weeks is the most cost-effective if you are limited by supply, but in practice it tends to be longer because often you have spare capacity.
Less sessions is a reliable way to reduce cost, but my understanding is there’s a U-shaped curve to cost-effectiveness here. 1 session doesn’t have enough benefits but 100 sessions costs too much and doesn’t add more benefit.
That sounds about right.
Also, are you targeting specific conditions? I see improvement in insomnia but that can arise from a sleep intervention or a general CBT course too
Depends on the client. Mostly our counselling is bespoke, but we have some programmes for more specialised issues (e.g. chronic insomnia, addiction, phobia)
I’m going to assume you mean comparison not experiment as we did no experiment comparing the two demographics.
The comparison was to show how much easier it is to treat high-functioning western demographics than it is to treat lower-functioning LMIC demographics. One common misconception I run into a lot is that treating people in LMICs is easier because there’s still “lower-hanging fruit” yet to be treated. I wanted to show some statistics illustrating that this was not the case by comparing two similar pilots with different demographics
The higher income, higher functioning demographic was easier to recruit, triage, maintain and got comparable results. I think this violates most funder’s expectations.
8 Counter-Intuitive Considerations for Psychotherapy Interventions
IMO, if the content is good we shouldn’t bring it up. If an author is producing bad content more than once a month and it seems generated by LLMs they should be warned then banned if it continues.
I suspect any comment threads about whether content is LLM-generated aren’t worth reading and thus aren’t worthwhile writing.
Example
https://upworkclone.bubbleapps.io/
Person specs
No technical knowledge required, it’d just involve learning how to use “bubble”. The hard part is that no volunteer wants to be on a platform with few organisations and no organisation wants to be on a platform with few volunteers. [further reading]
Do I wanna make this?
I have no time. Ideal founder traits:
1. Well connected with EA orgs volunteers find attractive or volunteers EA orgs would find attractive.
2. Experience volunteering
3. A love of networking
4. Good at convincing EAs to do stuff.
Unless do favors for, or otherwise suck up to, well-connected people it can be difficult to get your idea in front of early-stage funders (i.e. HNWIs / individual donors). Even if you do, unless you are able to get a warm introduction or a meeting, your idea will be a mere piece of paper, overlooked in favor of projects from people the funder already knows. Even if you manage to get connected, you will likely be rejected without feedback and so cannot efficiently improve your application.
Potential improvementsAI: I’m pretty sure a GPT 01-preview custom GPT, with just prompt engineering, could give pretty strong feedback on grant proposals if it was just given guidance and examples. The subset of applications that look strong to the bot could be forwarded to an email inbox where a well-connected EA could look at it and decide if it merits forwarding to funders.
Forum: There could be a weekly thread dedicated to feedback on ideas. Pay one person to give feedback to applications that get none.
EA YC: If either of the above yields promising results, it could be scaled up by doing an EA version of y-combinator.
A fuckton of EAs struggle with procrastination, at least 10% but probably more like 20-30%. Funders tend to underestimate the prevalence because it’s in nobody’s best interests to admit it to them.
Interventions work fast and are cheap. There’s no overhead because you can just do them via Zoom. We already know what works, so you don’t even have to innovate.
Funders I’ve spoken to have tended to think that it’s only the least productive EAs who procrastinate. I’ve worked with a bunch of people working in high prestige EA jobs, including dozens of charity founders, and can confirm this is categorically untrue. It’s not that highly productive people don’t procrastinate, it’s that they have other strengths that counteract the weakness.
A matchmaking service for EA projects and EA volunteers / freelancers with a reputation system: At present, volunteers are underutilised because there’s no good way of knowing who is reliable. Freelancers are often chosen because they are “EA-aligned” when really some are either incompetent or grifters. There ought be some way of seeing how someone has performed on past work.
Templates already exist on-line for this exact type of marketplace. The only upfront cost would be in finding enough early adopters to give it a go. From there, it could take a small cut of freelance work and fund itself.
The UK / US requirement seems both damaging and completely arbitrary. What’s the reasoning?Lots of strong charities, including many AIM-incubated ones, aren’t incorporated and instead work via a fiscal sponsor. Can you clarify if they are eligible?Edit: saw the footnote about UK/US.
Your work at BIT sounds really fantastic. Well done!
What could EA do better to help earn-to-givers maximise their impact?