Worked model
One model, opened up.
The research page shows how we measure in plain terms. This page shows the machinery for a single model, our analysis of long-lasting insecticidal nets for malaria prevention, so you can judge the work the way we do: assumption by assumption, source by source.
The program.
One million standard long-lasting insecticidal nets distributed in Mozambique through an established national distribution program, at a total cost of $4.85 million. Simple to describe. Not simple to value honestly.
The machine, end to end.
Below is the whole model as it actually runs, stage by stage. Open any stage to see the arithmetic underneath it – the same numbers we use, including the ones that make our results smaller.
29 countries modelled
6 benefit streams
3 readings
9 quality gates
01
What goes in
1,000,000 nets · $4.85M · one national distribution program
see the math
nets purchased and delivered
1,000,000
total program cost, delivery included
$4,850,000
÷
cost per net, on the ground
$4.85
No modelling yet – just the purchase. The modelling starts with what we refuse to count.
02
Where we shrink our own numbers
four conservative filters, every one pointing down
see the math
× 0.63
incrementality – we only count nets that would not have arrived anyway
−37%
protection life per net: 2.0 years, where the field assumes 2.5 or more
2.0 yr
÷
child-mortality effect: the Cochrane systematic review’s 17%, not the 50% assumed elsewhere
17%
If we err, we err against ourselves. Every stage below inherits these filters.
credit taken for behaviour-change campaigns running alongside the nets
$0
03
What comes out the other side
deaths, illness, and disability averted – then divided by cost
see the math
684
deaths averted
15,850
DALYs averted
239,626
cases averted
$4,850,000 across 684 deaths averted
$7,093 per life saved
published range from a leading independent evaluator
$3,000–$8,000
÷
$4,850,000 across 15,850 DALYs averted
$306 per DALY
Both headline numbers land inside their independent benchmarks – with all four filters from stage 02 still applied.
WHO cost-effectiveness threshold
1–3× GNI per capita
04
Six kinds of good, counted separately
no blended guess – each stream has its own valuation and evidence base
see the streams
Mortality
55.8%
Health system savings
18.0%
Household costs avoided
13.2%
Long-run development
9.2%
Morbidity
2.0%
Productivity
1.8%
Decomposition shown for the Democratic Republic of the Congo, as an example. Note what dominates: the value of a human life, counted directly.
05
One model, three readings
2.0:1 conservative · 12.5:1 method-aligned · 35.5:1 fully adjusted
see the math
our default: country-specific life-year value, floored at 20× GNI with 1.5 income elasticity – $347 per year in Mozambique
2.0:1
re-run with a major international benchmark’s group-average life-year value of $3,732
12.5:1
+
7.3 – remove the incrementality discount, as the benchmark does
19.8:1
The remaining gap is the benchmark’s non-linear transmission model, not a methodology error. We publish the 2.0:1.
+
9.9 – extend protection life to 3.0 years
29.7:1
+
5.8 – model 8 distribution rounds with 8% discounting and GDP growth
35.5:1
fully adjusted, against the benchmark’s published 46:1
77% match
06
Checked against the field, three ways
same model, three independent yardsticks
see the math
cost per life saved, against a leading evaluator’s published $3,000–$8,000
$7,093 – inside
cash-transfer multiple: our 6.0× against a published 9–23×; adopt their 50% mortality input and ours reads 19×
one assumption apart
×
29 countries – 29M nets, $140.65M: the model projects 24,372 deaths and 561,300 DALYs averted, at $5,771 per life and $251 per DALY
portfolio view
Three yardsticks, three explanations, zero unexplained gaps.
portfolio, fully adjusted, against the international benchmark’s published 48:1
50.1:1 – a 104% match
The disagreement, where it exists, is about how much value to assign – never about whether the intervention works. That is what honest modelling looks like: when we differ from the field, we show you exactly which assumption drives the difference, and what our number becomes under theirs.
And yes: every model we build and use gets this treatment.
Named sources.
This model stands on published work: the Cochrane systematic review of insecticide-treated nets; Shretta and Ngwafor (2023) for international malaria benefit-cost benchmarks; published cost-effectiveness ranges from independent charity evaluators; World Health Organization cost-effectiveness thresholds; and Robinson’s guidance on valuing statistical life years. The full citation list ships with the published methodology.
Want the next model when we publish it?