Disclaimer: This is my personal opinion and does not reflect the opinions of Giving What We Can or any particular other organisation. This is a draft.
Many people have heard that it costs about US$ 3,340 to save a life by distributing insecticide treated bed nets and ~$100 to enable a life of health i.e. $100 per Disability-adjusted life year (DALY) averted. Where do these numbers come from? How are they calculated? And how do they relate to each other?
These numbers are based on a calculation from Givewell’s review of the Against Malaria Foundation[1]. Givewell caveats that these estimates should be used with caution due to significant uncertainty around them, but that it is useful for comparative purposes and to “think through as many of the relevant issues as possible”[2].
Givewell’s spreadsheet[3] shows that they take the average overall costs for bednet distributions by AMF and divide them by the expected lives saved / death averted “per protected child under 5 – adjusted for today's lower rates of child mortality and insecticide resistance”. The later estimate also takes into account factors such as wastage, years of protection per person per net, and pre-existing net ownership. This calculation gives us the $3,340 per life saved / deaths averted figure.
But where does the $100 per DALY averted figure come from? And why does it cost $3,340 to avert a death of an 5-year-old child, which has a life expectancy of ~80 years, but ~$100 to enable a year of health life? Shouldn’t it be ~$40 per DALY averted (i.e. $3340 / 80 years)? The answer is that the $100 figure per DALY averted is time discounted at 3%. If one were to not time discount then it would indeed only be ~$40 per DALY averted. But if one time discounts ~80 DALYs at 3%, then they become ~30 DALYs (see [4] for the conversion table used and [5] to read more about time discounting). You might have noticed that 30*100 is $3000 and not $3,340. This is because the average DALY averted figure used is actually ~$112.5 (in Malawi it’s $98 per DALY averted, whereas in the DRC it’s $125 per DALY averted). 112.5*30 = 3375, which is pretty much the stated $3,340, with the discrepancy being due to rounding errors.
Incidentally, these figures are similar to the one published in a recent Lancet article[6], which estimates that, in low-income countries, to save a child's life, the cost is US$4205.
[5] Ord, Toby, and Robert Wiblin. "Should we discount future health benefits when considering cost-effectiveness?."
Casually, the Lancet article is making a very big claim - that general health spending on kids has led to averted deaths at $4205 each across the board since 2000. This is a simple average of the association between spending and reduced deaths. So it attributes 100% of the progress in this area to the amount of spending in each country. As they point out, the correlation is not tight. So it would not be sensible to attribute 100% of the improvement to the spending. Which is in a way disappointing because we can't then just throw more money at the problem and expect this return. But in another way isn't such a surprise that all Givewell found was a very slight improvement on that average. Education, culture, the economy, general efficiency of public services, declining fertility probably all had quite a lot to do with each country's story. So the Lancet article isn't that useful in this way?
The analysis is a bit more sophisticated than that:
"To avoid overestimation of the number of child deaths averted through health intervention, we take advantage of the analysis published by Wang et al (1), which examined the component of changes in child mortality that can be linked to changes in income per capita and educational attainment. Based on the econometric model used by Wang et al (1), we estimate counterfactual deaths in a scenario where income, education, and number of live births change with time but everything else is set at its year 2000 value." See supplemental material: http://www.thelancet.com/cms/attachment/2033982751/2049751458/mmc1.pdf
OK, so then either it is surprising that givewell hasn't found better, there are some harder to observe links between child health spending and child health, or there are other variables that probably aren't covered off properly in the model?
Big development organisations (Gates, UNICEF, etc) can do certain things that are very effective, such as vaccinations, and might also enjoy economies of scale. It's actually quite surprising that individual donors can do significantly better ($3340 per life saved by donating to AMF vs. $4205 per life saved). Note also, that the $4205 figure is a biased sample of any big development organisations programme - because most are not only active in low income countries. Individual donors cannot donate in a ringfenced fashion to development organisation through big organisations such as UNICEF unfortunately.