The Pollution–Development Trap

The first loop runs from poverty to exposure. Households that cannot afford gas or electricity cook on wood, charcoal, dung or kerosene, filling their homes and neighbourhoods with smoke: 84% of people in low-income countries do so, and 2.1 billion people worldwide. Cities that cannot afford reliable power run on diesel generators; fleets of second-hand vehicles without filters crowd unpaved roads; waste and crop residues are burned in the open. Regulators lack the budgets, the monitors and the legal powers to enforce standards — many governments in Africa publish no air-quality data at all. And the early stage of industrialisation, built on coal, brick kilns and heavy industry, adds pollution faster than rising incomes create the demand and the means to control it.

The second loop runs from exposure back to income, and here the evidence comes from natural experiments that isolate pollution from everything else. Ebenstein and colleagues used China's Huai River heating policy to show that sustained exposure shortens lives, a result the Air Quality Life Index expresses as 0.98 years of life expectancy per 10 µg/m³ of PM2.5. In Mexico City, the closure of a refinery cut sulphur dioxide nearby by 19.7% and raised hours worked by 3.5%. Across Chinese manufacturing, each 1 µg/m³ less PM2.5 raised firm productivity by 0.82%; across European regions, each 1 µg/m³ more cut GDP by 0.8%, almost entirely through output per worker. In Israel, pollution on exam days lowered scores, university attainment and later earnings.

Added up, the costs are large. The World Bank put the health cost of PM2.5 at $8.1 trillion in 2019, 6.1% of world GDP, with most of the burden in low- and middle-income countries. In India, lost output from air-pollution deaths and illness alone was $36.8 billion in 2019, 1.36% of GDP. Lead — still present in batteries, paint, cookware and spices — cost an estimated $6.0 trillion, 6.9% of world GDP, through lost IQ and heart disease, with 95% of the IQ loss in low- and middle-income countries. These are costs that fall on the workers and children on whom future growth depends.

Weighted by population, outdoor PM2.5 in 2024 is highest in lower-middle-income countries, at 42.6 µg/m³, against 30.0 in low-income, 26.0 in upper-middle-income and 12.5 in high-income countries. That is the hump: the countries now industrialising, led by India, Pakistan, Bangladesh, Nigeria and Egypt, breathe the dirtiest outdoor air. Applying the Air Quality Life Index's coefficient, PM2.5 above the WHO guideline costs 3.7 years of life expectancy in lower-middle-income countries, 2.4 in low-income, 2.1 in upper-middle-income and 0.7 in high-income countries — about 2.5 years for the average person, or 20 billion years of life across the 8.1 billion people covered.

Household air pollution reverses the order. In low-income countries 84% of people cook with polluting fuels, against 36% in lower-middle-income, 10% in upper-middle-income and almost none in high-income countries. Combining both sources, the World Health Organization's age-standardised death rate from air pollution is 186 per 100,000 in low-income countries, 148 in lower-middle-income, 84 in upper-middle-income and 25 in high-income countries — a ratio of 7.6 between the poorest and richest.

The World Bank's adjusted-savings accounts value the damage from particulate pollution at 1.6% of national income in low-income countries and 1.3% in lower-middle-income ones, against 0.1% in high-income countries; this measures forgone output, not the far larger welfare cost of early death. The same ladder runs through human capital: the average Human Capital Index is 0.37 in low-income and 0.73 in high-income countries, stunting 37% against 3%, and learning poverty — ten-year-olds unable to read a simple text — 91% against 9%. Pollution is one cause among many of these gaps, as section 5 discusses.

Each country's path since 1998 is classified by its three-year average PM2.5. Countries that have cleaned up peaked at 15 µg/m³ or more and have since fallen by at least a quarter to below 20 — 22 countries, mostly in Europe, plus Malaysia, Singapore and Hong Kong. Countries that are turning peaked above 25 before 2020 and have fallen at least 15% but remain above 20: China, India, Nepal and a handful of others, holding 3.0 billion people. Stuck countries remain above 25 µg/m³ without a clear decline — 36 countries and 1.7 billion people, including Pakistan, Nigeria, Bangladesh, Egypt, the DR Congo, Viet Nam and Iran. Rising countries are above 15–20 and still climbing: 30 countries and 0.8 billion people, from Iraq and Afghanistan to Kenya, Angola, Zambia, the Philippines and South Africa.

China is the defining case. Its PM2.5 rose to a three-year peak of 50 µg/m³ in 2014, when income per head was about $14,000 at purchasing-power parity, and then fell by 5.5% a year to 31 in 2024 after the 2013 Air Pollution Prevention and Control Action Plan targeted coal, heavy industry and regional coordination together. India peaked at 58 in 2017, at about $7,100 a head, and stood at 44 in 2024; part of that fall reflects weather and the pandemic, and it remains nine times the WHO guideline. Among countries with a peak between 2001 and 2019, the median income at the peak was about $24,000 — but the most recent turns have come at far lower incomes.

A caution on the poorest countries: several in the Sahel and West Africa appear as turning or stuck mainly because of desert dust and seasonal fires, which swing with the weather and which policy cannot easily control. Their trends say less about development than China's or Poland's. The trap set used in this report is therefore defined by persistence — stuck or rising — together with low income: 35 countries with 1.45 billion people where PM2.5 has not turned down and GDP per head is below $10,000.

If pollution were simply a phase that growth outgrows, faster-growing countries should have seen their air improve faster. The chart tests this by splitting countries into thirds by their income in 2000 and comparing each country's growth in income per head from 2000 to 2024 with the change in its PM2.5 from 2005 to 2024.

In the richest third, with incomes above about $14,900 in 2000, PM2.5 fell by 0.9% a year on average, and faster growth went with faster improvement: each extra point of annual growth was associated with a further 0.28-point annual decline in PM2.5. In the middle third the association was similar (0.19) but the average level barely moved. In the poorest third — incomes from $700 to $4,800 — PM2.5 rose by 0.15% a year on average, and there was no association at all between growth and air quality (correlation −0.01), even though these countries grew by 2.4% a year per head, as fast as the middle third.

This is the trap in its simplest form. Growth at low incomes adds vehicles, generators, industry and urban sprawl faster than it brings the regulation, cleaner fuels and monitoring that bend pollution down. Waiting for growth to fix the air means waiting for decades of exposure that themselves slow growth. Countries that turned early, like China, did so through deliberate policy rather than income alone.

The World Bank's Human Capital Index measures the productivity a child born today can expect to reach as a worker, given the health and education she is likely to receive; a score of 0.5 means half of the potential under full health and education. Across 144 countries, PM2.5 is strongly associated with a lower score even after accounting for income: each 10 µg/m³ more PM2.5 goes with a score about 0.022 lower (t = −7.2). With controls for the seven world regions the association shrinks only slightly, to 0.019 (t = −4.7). For comparison, the gap between the average lower-middle-income and high-income country is about 0.26.

Stunting tells a similar story. At the same income, each 10 µg/m³ more PM2.5 goes with 1.3 percentage points more stunted children (t = 2.6; 1.2 with regional controls), and each 10 percentage points more of the population cooking with polluting fuels with 1.1 points more stunting (t = 3.0), though the latter weakens with regional controls. Learning poverty is associated with pollution across countries, but not within regions.

These are cross-country associations, and pollution travels with other disadvantages — crowding, poor sanitation, disease, weak schools — that income alone does not capture. They are best read alongside the causal studies: smoke from cooking fires and crop burning has been linked to lower birth weight and child growth, pollution on exam days lowers scores and later earnings, and lead exposure, often concentrated in the same poor neighbourhoods, permanently lowers IQ. The direction of every finding is the same: pollution erodes the human capital on which long-run growth depends.

Household air pollution is the most direct link between poverty and exposure: people burn wood, charcoal and dung because they cannot afford or obtain cleaner fuels, and the smoke falls hardest on women and young children. Worldwide, access to clean cooking rose from 48% in 2000 to 75% in 2024. The Tracking SDG7 report counts 2.1 billion people still cooking with polluting fuels in 2023, split almost equally between Asia and Africa.

The two regions are moving at very different speeds. South Asia's access rose from 20% in 2000 to 74% in 2024, driven by India's subsidised LPG connections, and at its 2015–2024 pace of 3.2 points a year it would reach 95% around 2030. East Asia went from 40% to 87%. Sub-Saharan Africa rose from 9% to 23%, gaining less than a point a year since 2015; at that pace it would not reach 95% until around 2103, and the number of people without access there is still growing with population.

Because this half of the loop is driven by poverty itself, it is where policy has the clearest leverage. Clean cooking does not require a country to get rich first: it requires fuel supply chains, connection subsidies, results-based finance and, increasingly, efficient electric cooking on the grids and mini-grids now being built. Whether new stoves and connections are actually used — affordability of refills, reliable supply, appliances that suit local cooking, and honest measurement — matters as much as how many are handed out.

The trajectory layer colours each country by its path since 1998. The countries stuck high or rising form a nearly continuous belt from Senegal and Nigeria through the Sahel, Sudan, Egypt and the Middle East to Afghanistan, Pakistan, Bangladesh and Myanmar, with outliers in East and southern Africa and the Andes; within it, the low-income ones make up the trap set. Cleaned-up countries are concentrated in Europe; China and India are turning.

The exposure and life-expectancy layers show the levels behind those paths: the deepest losses — four to seven years of life expectancy — are in Qatar, Bangladesh, Kuwait, the United Arab Emirates, Pakistan, Cameroon, Nigeria and Saudi Arabia, where combustion, desert dust or both are high. The cooking-fuel layer shows the household half of the loop, overwhelmingly African, and the damage layer the World Bank's estimate of the particulate damage in each economy.

Hover over a country to see its card: its trajectory class, PM2.5 now and at its peak, the trend since 2014, income per head now and at the peak, cooking fuels, the death rate, damage, Human Capital Index, stunting and life expectancy lost, with a sparkline of PM2.5 since 1998. Click a country to load it into the simulator below.

Choose a country and an annual rate of decline in PM2.5 from 2024 to 2050. China's rate since 2014 — 5.5% a year — is the default; the country's own trend since 2014 is offered for comparison. The simulator projects PM2.5 to 2050, marks when it would cross the WHO's interim targets of 35, 25, 15 and 10 µg/m³, and estimates two gains relative to holding PM2.5 at its 2024 level.

The first is life expectancy, using the Air Quality Life Index's coefficient of 0.98 years per 10 µg/m³, applied only to concentrations above the WHO guideline of 5. The second is output, using a constant elasticity of GDP to PM2.5: 0.039, the nationwide general-equilibrium estimate of Fu, Viard and Zhang for China, as a conservative bound, and about 0.08, the effect implied by Dechezleprêtre, Rivers and Stadler for European regions, as an upper bound.

These are illustrative long-run gains, not forecasts. They assume that the causal estimates from China and Europe apply elsewhere, that the full decline in PM2.5 comes from controllable sources — which overstates what is possible where desert dust dominates — and they leave out household air pollution, which falls with clean cooking rather than with outdoor controls, and the gains to children's human capital.

The first lesson of the data is that countries need not wait. China began cutting pollution at about half the income at which most of Europe did, and India appears to have started at half of China's. The technologies that make this possible — flue-gas controls, cleaner vehicle standards and fuels, LPG and electric cooking, cheap solar power, low-cost sensors and satellite monitoring — are far cheaper than when today's rich countries cleaned up. Air-quality management belongs in growth strategies and budgets from the start, financed by the productivity and health it returns.

The second is to attack the poverty-driven sources directly. Clean cooking needs connection subsidies, fuel supply and results-based finance, and is where sub-Saharan Africa is falling furthest behind. Open burning of waste and crop residues, diesel generators that stand in for unreliable grids, and old, unfiltered vehicles each have known remedies. Lead is the clearest case: the end of leaded petrol, completed when Algeria stopped selling it in July 2021, is estimated to prevent more than 1.2 million premature deaths and save $2.45 trillion a year; the lead that remains in battery recycling, paint, cookware and spices is a cheap target with large gains for children's intelligence.

The third is to see the problem. Many of the countries in the trap have no government monitoring and no open data, so pollution is invisible in policy debates; in 2025 the end of the US State Department's overseas monitoring programme left 36 countries without their only high-quality source of air-quality data, and Africa received less than 1% of philanthropic funding for outdoor air quality between 2019 and 2023. Monitoring is cheap relative to the losses it reveals. Finally, policy must be regional: pollution crosses borders and city limits, as the companion report on atmospheric transport shows, and in the Sahel natural dust sets a floor that only adaptation — filtration, early warning, health services — can address.

Coverage: 159 countries with at least 1 million people and a full PM2.5 series. PM2.5: population-weighted satellite-based estimates (SatPM, Washington University in St. Louis, via Our World in Data), 1998–2024, smoothed as three-year averages to identify peaks. Income: GDP per head at purchasing-power parity in constant 2021 international dollars (World Bank WDI). Trajectories: cleaned up = peak ≥ 15 µg/m³ and current at least 25% below the peak and below 20; turning = peak ≥ 25 before 2020, current at least 15% below the peak and ≥ 20; rising = current ≥ 20 with a 2014–2024 trend above +0.5% a year, or ≥ 15 with a trend above +1%; stuck = current ≥ 25 otherwise; low = current below 15; middle = the rest. Trap set: stuck or rising with GDP per head below $10,000.

Burden: polluting cooking fuels = 100 minus access to clean fuels and technologies (WHO via WDI, latest year); death rate = WHO age-standardised mortality attributed to household and ambient air pollution, 2019; damage = adjusted savings: particulate emission damage, % of GNI, 2021; Human Capital Index 2020; stunting and learning poverty, latest year. Life expectancy lost = 0.098 × (PM2.5 − 5), the Air Quality Life Index coefficient. Growth test: annual log growth of GDP per head 2000–2024 against annual log change in PM2.5 2005–2024, by third of 2000 income. Partial associations: OLS with log GDP per head (and, where stated, World Bank region fixed effects), heteroskedasticity-robust t-statistics.

Caveats: satellite PM2.5 includes natural dust; cross-country associations are not causal; the AQLI and output elasticities come from China, Europe and Mexico; WHO mortality and the damage series use different methods and years from SatPM; income groups use the World Bank's current classification. Related reports: The Unequal Geography of Human Air Pollution Exposure, The Geography of Human Pollution Exposure and the pollution–poverty–climate nexus.