This is a great study and I’m surprised and impressed that morality dropped this much. This is one of the first RCT studies in a while (I think) to show such a clear mortality drop. After considering the benefits cash helps with such as safe delivery, reduced malnutrition and reduced maternal work—it makes a lot more sense.
One caveat to keep in mind is that this program cost around 30 million dollars and counterfactually saved about 90 infant lives. So that’s about $300,000 per life saved. So cash is not a cost-effective intervention for saving lives alone (and it’s not claiming to be).
Obviously they would have saved a lot more lives than that through reduced maternal mortality and under 5 mortality as well, through similar mechanisms. Maybe they’ll measure that in future too....
I do wish other interventions had the opportunity to study these kind of things like cash does. Other interventions just don’t get the great opportunities cash transfers do to probe all the potential benefits around the edges. It’s pretty cool.
Still, has updated me on the effectiveness of cash that’s for sure.
They do point out that the 30 million dollars was spread out among everyone, not just pregnant women. They take a guess at what the cost per life saved would be if it was targeted specifically at pregnant women:
targeting UCTs to women in the third trimester of pregnancy under these assumptions would cost about USD PPP 92,000 (or $39,000 in nominal dollars) per child death averted.
We should get more data on the actual cost-effectiveness in a while from the targeted givedirectly work.
100 percent agree. For sure the 30 million dollars is non-targeted (the child mortality benefit might not even make the top 3 or so benefits of cash). And yes the new study should give us more insight into the specific benefits of cash on child mortality. If anything I think the effect size will likely be smaller this study − 50% is a pretty insane start.
Yes, it’s important to take into account that this is the finding of one study, whereas the mosquito net results come from a much more rigorous cochrane metastudy of many different studies.
Do you have more reasons to be skeptical of the 47% figure? After all, with 1000 bucks the household would be able to buy all the other interventions.
I’m only skeptical on priors, because 50 percent is an unusually huge infant mortality drop. I can’t off the top of my head think of another right now like that measured in the last few years. I agree it feels plausible though and like your say 1000 dollars buys as lot. In the public health field we hardly ever see those kind of numbers even for expensive targeted interventions.
It’s reasonable to be somewhat skeptical based on priors given the statistical power of this (very worthy and interesting!) study? I didn’t dig deeper, but back of the envelope if you draw from 10000 iid households with an infant each and a 4% probability you’d expect a standard deviation around 0.2%, so there’s not much room for slicing the data a lot finer or additional correlation creeping in without a decent amount of sampling error. Obviously, with smarter analysis you can do a bit better and it’s hard and expensive to get more data, but it’s easy to believe the results are biased upwards a bit. The study is a great step in the right direction.
It’s “a cost of USD PPP 299,418 per death averted,” which is about $125,000 nominal USD based on the conversion implied by other parts of the article. At least to the extent that people are comparing to GiveWell estimates—and I suspect most readers will be—the nominal figure may be the better figure to highlight here.
While the broad cash program in this study is certainly more expensive per family than other global health programs, the researchers note that if the cash was targeted to pregnant women in their third trimester, it could be “comparably cost effective to a number of WHO-recommended maternal and child health interventions, even without taking into account other possible benefits of unconditional cash transfers (such as consumption gains).”
We’re launching a pilot this fall in Kenya specifically for pregnant women to learn just how much more cost-effectively cash can improve infant and maternal health, one of our many tests to improve our programs for specific outcomes.
Targeting UCTs to women in the third trimester of pregnancy under these assumptions would cost about USD PPP 92,000 (or $39,000 in nominal dollars) per child death averted. We can benchmark these calculations to 37 WHO-recommended maternal and child health interventions in East Africa as estimated by Stenberg et al. (2021). Across interventions and scenarios, the cost per death averted ranges from USD PPP 27 to USD PPP 222,952.[1] Hence, even without taking into account any of the other documented benefits of UCTs (such as gains in consumption), the transfers are squarely in the range of cost per death averted among these WHO-recommended interventions.
Article at p. 34 (footnote # is 36).
Stenberg et al. is here. Eastern sub-Saharan Africa is table 3. I’m not sure how to convert HLYs into deaths averted, but of the 37 interventions, #36 (ACER [2]of 1156.2) and #37 (ACER of 1310.6) are significantly less cost-effective than even #35 (ACER of 355.9). Based on the range in the article, it sounds like UCT-for-pregnant-women might rank somewhere between #35 and #36 here?
That doesn’t sound like a particularly strong showing. The 27th out of 37 interventions has an ACER of 94.9. Some interventions I’ve seen discussed in EA circles are rated by Stenberg et. al much lower than even that: Vitamin A supplementation (0-4 years), 7.1; Kangaroo mother care, 20.1; Syphilis detection and treatment in pregnancy, 24.8.
Stenberg et al. (2021) evaluates cost-effectiveness using three coverage level scenarios: 50%, 80%, and 95%, and report health impacts in terms of healthy life years (HLY) saved. We converted HLYs to deaths averted using WHO data on total and healthy life expectancy in Kenya (World Health Organization, 2025).
I imagine that’s the purpose of the trial—to optimise the programme for lifesaving and get that cost-per-life down.
They’ve already got data that villages nearer a physician show stronger benefit, timing on month of birth shows stronger benefit. I wouldn’t be surprised to find other things (e.g. maternal age, child number, season in year, country, harvest quality) contributing and therefore optimisable, as well as the benefit/$ rising if the $ amount drops.
Also the 77% drop from a one-time intervention is already a huge whole-issue tackle that’s going to have massive donor appeal (e.g. give $1000 to safeguard one child—this child). We had a Forum post some time back from someone expressing strong preference for knowing exactly where their money went. It’s a very common donor preference.
Yes thanks for the reply. For sure cash here meets a WHO bar, as do a lot of health interventions. I used to quote the WHO bars a bunch but I’m not sure how useful they are practically as so many interventions meet those bars that we can’t realistically fund them all.
I was implicitly considering cost effectiveness compared to GiveWell and open Philanthropy bars (as we often do here on the forum) which are a lot higher than the WHO’s.
Really looking forward to the pilot in Kenya great job GiveDirectly team!
‘even without taking into account other possible benefits of unconditional cash transfers (such as consumption gains)’
Presumably we’ll know from the study when it comes out, which will be exciting, but if mother’s are spending the money on services related to making their pregnancy/delivery safer, will they also receive the same amount of consumption gains? - I’d think these more general gains would be lower unless the pregnancy related costs are a lot less than the value of the transfer.
This is a great study and I’m surprised and impressed that morality dropped this much. This is one of the first RCT studies in a while (I think) to show such a clear mortality drop. After considering the benefits cash helps with such as safe delivery, reduced malnutrition and reduced maternal work—it makes a lot more sense.
One caveat to keep in mind is that this program cost around 30 million dollars and counterfactually saved about 90 infant lives. So that’s about $300,000 per life saved. So cash is not a cost-effective intervention for saving lives alone (and it’s not claiming to be).
Obviously they would have saved a lot more lives than that through reduced maternal mortality and under 5 mortality as well, through similar mechanisms. Maybe they’ll measure that in future too....
I do wish other interventions had the opportunity to study these kind of things like cash does. Other interventions just don’t get the great opportunities cash transfers do to probe all the potential benefits around the edges. It’s pretty cool.
Still, has updated me on the effectiveness of cash that’s for sure.
They do point out that the 30 million dollars was spread out among everyone, not just pregnant women. They take a guess at what the cost per life saved would be if it was targeted specifically at pregnant women:
We should get more data on the actual cost-effectiveness in a while from the targeted givedirectly work.
100 percent agree. For sure the 30 million dollars is non-targeted (the child mortality benefit might not even make the top 3 or so benefits of cash). And yes the new study should give us more insight into the specific benefits of cash on child mortality. If anything I think the effect size will likely be smaller this study − 50% is a pretty insane start.
Yes, it’s important to take into account that this is the finding of one study, whereas the mosquito net results come from a much more rigorous cochrane metastudy of many different studies.
Do you have more reasons to be skeptical of the 47% figure? After all, with 1000 bucks the household would be able to buy all the other interventions.
I’m only skeptical on priors, because 50 percent is an unusually huge infant mortality drop. I can’t off the top of my head think of another right now like that measured in the last few years. I agree it feels plausible though and like your say 1000 dollars buys as lot. In the public health field we hardly ever see those kind of numbers even for expensive targeted interventions.
It’s reasonable to be somewhat skeptical based on priors given the statistical power of this (very worthy and interesting!) study? I didn’t dig deeper, but back of the envelope if you draw from 10000 iid households with an infant each and a 4% probability you’d expect a standard deviation around 0.2%, so there’s not much room for slicing the data a lot finer or additional correlation creeping in without a decent amount of sampling error. Obviously, with smarter analysis you can do a bit better and it’s hard and expensive to get more data, but it’s easy to believe the results are biased upwards a bit. The study is a great step in the right direction.
It’s “a cost of USD PPP 299,418 per death averted,” which is about $125,000 nominal USD based on the conversion implied by other parts of the article. At least to the extent that people are comparing to GiveWell estimates—and I suspect most readers will be—the nominal figure may be the better figure to highlight here.
While the broad cash program in this study is certainly more expensive per family than other global health programs, the researchers note that if the cash was targeted to pregnant women in their third trimester, it could be “comparably cost effective to a number of WHO-recommended maternal and child health interventions, even without taking into account other possible benefits of unconditional cash transfers (such as consumption gains).”
We’re launching a pilot this fall in Kenya specifically for pregnant women to learn just how much more cost-effectively cash can improve infant and maternal health, one of our many tests to improve our programs for specific outcomes.
(updated our post above to clarify this)
Adding this quote for context:
Targeting UCTs to women in the third trimester of pregnancy under these assumptions
would cost about USD PPP 92,000 (or $39,000 in nominal dollars) per child death averted.
We can benchmark these calculations to 37 WHO-recommended maternal and child health
interventions in East Africa as estimated by Stenberg et al. (2021). Across interventions
and scenarios, the cost per death averted ranges from USD PPP 27 to USD PPP 222,952.[1]
Hence, even without taking into account any of the other documented benefits of UCTs (such
as gains in consumption), the transfers are squarely in the range of cost per death averted
among these WHO-recommended interventions.
Article at p. 34 (footnote # is 36).
Stenberg et al. is here. Eastern sub-Saharan Africa is table 3. I’m not sure how to convert HLYs into deaths averted, but of the 37 interventions, #36 (ACER [2]of 1156.2) and #37 (ACER of 1310.6) are significantly less cost-effective than even #35 (ACER of 355.9). Based on the range in the article, it sounds like UCT-for-pregnant-women might rank somewhere between #35 and #36 here?
That doesn’t sound like a particularly strong showing. The 27th out of 37 interventions has an ACER of 94.9. Some interventions I’ve seen discussed in EA circles are rated by Stenberg et. al much lower than even that: Vitamin A supplementation (0-4 years), 7.1; Kangaroo mother care, 20.1; Syphilis detection and treatment in pregnancy, 24.8.
Stenberg et al. (2021) evaluates cost-effectiveness using three coverage level scenarios: 50%, 80%, and 95%, and report health impacts in terms of healthy life years (HLY) saved. We converted HLYs to deaths averted using WHO data on total and healthy life expectancy in Kenya (World Health Organization, 2025).
“The average cost-effectiveness ratios (ACERs) were calculated by dividing the total cost for scale-up by the total health gain.”
I imagine that’s the purpose of the trial—to optimise the programme for lifesaving and get that cost-per-life down.
They’ve already got data that villages nearer a physician show stronger benefit, timing on month of birth shows stronger benefit. I wouldn’t be surprised to find other things (e.g. maternal age, child number, season in year, country, harvest quality) contributing and therefore optimisable, as well as the benefit/$ rising if the $ amount drops.
Also the 77% drop from a one-time intervention is already a huge whole-issue tackle that’s going to have massive donor appeal (e.g. give $1000 to safeguard one child—this child). We had a Forum post some time back from someone expressing strong preference for knowing exactly where their money went. It’s a very common donor preference.
Well done GiveDirectly!
Yes thanks for the reply. For sure cash here meets a WHO bar, as do a lot of health interventions. I used to quote the WHO bars a bunch but I’m not sure how useful they are practically as so many interventions meet those bars that we can’t realistically fund them all.
I was implicitly considering cost effectiveness compared to GiveWell and open Philanthropy bars (as we often do here on the forum) which are a lot higher than the WHO’s.
Really looking forward to the pilot in Kenya great job GiveDirectly team!
‘even without taking into account other possible benefits of unconditional cash transfers (such as consumption gains)’
Presumably we’ll know from the study when it comes out, which will be exciting, but if mother’s are spending the money on services related to making their pregnancy/delivery safer, will they also receive the same amount of consumption gains? - I’d think these more general gains would be lower unless the pregnancy related costs are a lot less than the value of the transfer.