Ruth makes this point, rightly, every year.
By this stage, @GiveWell's recs start to look more like a persistent bias or limitation inherent in their model, rather than a credibly objective take on different charities' effectiveness.
(Thread incoming)
By this stage, @GiveWell's recs start to look more like a persistent bias or limitation inherent in their model, rather than a credibly objective take on different charities' effectiveness.
(Thread incoming)
Their focus on where a next marginal donor dollar can do the most good is appealing, in theory.
But it presumes that's a useful way to measure effectiveness.
In practice this = heavy bias for simple, linear interventions that perform similarly across diverse environments.
But it presumes that's a useful way to measure effectiveness.
In practice this = heavy bias for simple, linear interventions that perform similarly across diverse environments.
That describes a subset of dev projects that are amenable to linear, quantitative measurement - mostly health-related. The model consistently rewards them.
But plenty of dev projects - indeed most! - aren't linear. Particularly for @GiveWell's other target - poverty alleviation.
But plenty of dev projects - indeed most! - aren't linear. Particularly for @GiveWell's other target - poverty alleviation.
Unsurprising then that @GiveWell's model struggles to find non-global health charities to recommend, outside of @Give_Directly, whose model (give cash!) is simple to measure.
So either there is no other forms of poverty alleviation worth supporting, or the eval model is flawed.
So either there is no other forms of poverty alleviation worth supporting, or the eval model is flawed.
My money (literally, in terms of where I donate) is on the latter.
Poverty alleviation (and much health work, too) is not a linear process built upon vertical interventions. It is a complex, context-dependent, non-linear process influenced heavily by exogenous factors.
Poverty alleviation (and much health work, too) is not a linear process built upon vertical interventions. It is a complex, context-dependent, non-linear process influenced heavily by exogenous factors.
Indeed there is a robust school of thought in dev practice around non-linear, adaptive interventions that embrace the complexity and diversity of fragile and developing environments, rather than seeking universal solutions. odi.org/projects/2918-…
I suspect that @GiveWell is sophisticated enough to recognize this. They're smart folks and their model is pretty interesting.
But I wish they would talk about it in a humbler and more nuanced way, rather than promoting it as the best way to donate to charity.
But I wish they would talk about it in a humbler and more nuanced way, rather than promoting it as the best way to donate to charity.
A more accurate way to understand their model is that it's basically the SAT. Is the SAT the final word on student potential? Not remotely. It has inadvertent biases and blind spots, and rewards a narrow kind of performance. insidehighered.com/news/2010/06/2…
There is value in seeing someone's SAT score - within a wider context of their overall academic potential. But making college admission strictly a matter of SAT performance would be crazy.
But this is what the @GiveWell model rewards: a particular subset of activities that produce what the model can measure, while marginalizing the rest.
Like a college admitting only 1600-SAT students from privileged private schools.
Like a college admitting only 1600-SAT students from privileged private schools.
In sum, then - by all means have a look at @GiveWell's recs. But don't mistake them for the final word on charity effectiveness.
Recognize them instead as one useful perspective on the most cost-effective vertical, linear-impact, mostly-global-health programs.
(rant over)
Recognize them instead as one useful perspective on the most cost-effective vertical, linear-impact, mostly-global-health programs.
(rant over)
The shape of the eval tool will shape how orgs prioritize. So just as colleges have adapted their priorities (and sometimes gamed their stats) to rise in the US News rankings, a one-size-fits-all impact value/impact model could skew charity priorities. theatlantic.com/education/arch…
How? Well as @spcharle and @caitlin_tulloch point out, operations in fragile settings are inherently more expensive than in stable settings. The model's rankings would punish those operations, because it means higher per-beneficiary cost.
The ethics of saving only the cheap-to-save lives are...pretty questionable.
But it's also shortsighted because most of the development gap in coming decades will be in these more-difficult settings, not in stable developing countries.
But it's also shortsighted because most of the development gap in coming decades will be in these more-difficult settings, not in stable developing countries.
Further, as @RChandran1 points out, broad policy reform and institution-building investments can have HUGE impact. But are also higher risk and take longer to reach impact.
And that raises a really critical attributability issue. A vax or deworming intervention will have direct and reliable impact attribution at a definable cost. But many/most dev outcomes are multi-variable and their impacts will be attributable only partially to the intervention.
So the @GiveWell model writes off organizations that do that kind of more complex, long-term, multi-variable work (e.g. how dev't actually happens).
The more I think about it, the more I see parallels here to a very old debate in charity-land: child sponsorship. Stick with me.
The more I think about it, the more I see parallels here to a very old debate in charity-land: child sponsorship. Stick with me.
Child sponsorship was a popular fundraising tool for years because it provided an appearance of certainty to prospective donors - *this kid*, on your fridge magnet, is being specifically helped by your money.
@Kiva's initial model used a similar pitch, but for microcredit.
@Kiva's initial model used a similar pitch, but for microcredit.
But these models came in for critiques.
Child sponsorship is pretty messy on the ground and frequently doesn't go to the kid on your magnet: nytimes.com/2016/08/03/wor…
And Kiva wasn't sending your loan direct to the entrepreneur whose story you clicked on: cgdev.org/blog/kiva-not-…
Child sponsorship is pretty messy on the ground and frequently doesn't go to the kid on your magnet: nytimes.com/2016/08/03/wor…
And Kiva wasn't sending your loan direct to the entrepreneur whose story you clicked on: cgdev.org/blog/kiva-not-…
What these models did was foster a sense of connection and certainty between donor and recipient - albeit an overblown or misleading connection.
@GiveWell is somewhat a modern-day spin on the same concept.
@GiveWell is somewhat a modern-day spin on the same concept.
Instead of the debunked approach of "your $20 can buy a goat for the Diaby family", we get "your $20 can keep X kids from getting malaria".
It's still selling the same misleading sense of certainty, albeit wrapped up in (much) better data and analysis.
It's still selling the same misleading sense of certainty, albeit wrapped up in (much) better data and analysis.
Does your money keep some kids from getting malaria? Yup.
But are those kids necessarily the most-at-risk? Is malaria the biggest dev obstacle they face? Is it their own biggest priority?
Hard to say. And answering those Qs costs more money than just providing a bednet.
But are those kids necessarily the most-at-risk? Is malaria the biggest dev obstacle they face? Is it their own biggest priority?
Hard to say. And answering those Qs costs more money than just providing a bednet.
So look. Data is a good thing, but it's not the only thing. Judgment and context are as/more important to enduring dev outcomes. But they add cost and are hard to quantify.
And @GiveWell's model, like an old-school child sponsorship program, offers a silver bullet option (albeit dressed up in contemporary data analytics) in the face of the messy, complex reality of development practice.
If that's what you're looking for with your money, go for it.
If that's what you're looking for with your money, go for it.
Relatedly several folks have asked me where I donate. I typically give to orgs whose fieldwork I've seen in action, and whose people I know to be smart and evidence-focused while contending with extremely tough environments.
There are many such organizations; this list is not exhaustive. But here are some that I support:
@MSF (impressive frontline healthcare even tho I sometimes differ with their advocacy approach)
@RESCUEorg (one of the best emergency NGOs, great GBV programs and refugee advocacy)
@MSF (impressive frontline healthcare even tho I sometimes differ with their advocacy approach)
@RESCUEorg (one of the best emergency NGOs, great GBV programs and refugee advocacy)
A few more:
@mercycorps (smart, thoughtful approach to applied evidence in fragile settings)
@hrw (amazing, expeditionary, impactful human rights advocacy)
And usually mix in a few others that have done impressive things in a given year (e.g. @SavetheChildren in Puerto Rico)
@mercycorps (smart, thoughtful approach to applied evidence in fragile settings)
@hrw (amazing, expeditionary, impactful human rights advocacy)
And usually mix in a few others that have done impressive things in a given year (e.g. @SavetheChildren in Puerto Rico)
OK, now I'm finally done. Thanks to those who stuck it out to the end.
