What we asked: which countries have traded their way into a starch-rich, micronutrient-poor food supply, and does it show up in women’s anemia?
We spent three weeks trying to blame food trade for that. It wasn't trade. This is the model that found what it actually was, and you can operate it yourself.
This is the regression from our analysis, running live. It learned from 167 countries. Give it a diet and an income and it estimates how many women aged 15 to 49 would be anemic there.
Start with a real country, then change one thing at a time. The interesting part isn't the number it gives you. It's which sliders move it, and which ones barely do anything at all.
Load an actual country's numbers so you have a baseline you can trust before you start changing things.
Share of daily calories from cereals, roots and tubers. Low means a varied plate; high means one or two staples.
Share of the food supply bought from abroad. This was our original hypothesis. Move it and watch the number.
GDP per person, adjusted for local prices. We held it still to see whether diet mattered on its own.
Horizontal: how much of the diet is starch. Vertical: how many women are anemic. Click any dot to load that country into the model.
Pooled across 178 national food supplies, latest year. Home-grown calories win on four of five micronutrients. Only B12 goes the other way, because meat and dairy travel too.
Why. Anemia in women of reproductive age is common, and it's getting worse. It rose in 123 of 162 countries since 2010. Two reasons this is a food-supply problem, not just a health statistic. One: the data measures food available per person, nationally. That average can look fine on paper while women don't get their share of it. They're more food-insecure than men in 98 of 135 countries. Two: the average doesn't adjust for biology. Iron needs peak during the exact reproductive years this analysis covers.
What. Which countries have traded their way into a starch-rich, micronutrient-poor food supply, and does it show up in women's anemia? The reasoning: starch, sugar and oil ship and store well; vegetables, eggs and fish don't, so trade quietly selects for calories over nutrition without anyone deciding it. That's a clean, testable chain. We tested it four ways. It broke.
Who. 161–167 countries with complete records (178 in the raw data; some of the worst-affected are missing entirely; see the assumptions below), looking at women aged 15–49. Built by four of us: Mansi Modi, Amia Nagpal, Gargee Nimdeo, Alexandra Harlan.
How. Public data: FAOSTAT, World Bank, WHO/DHS. A regression you can run live on the Model tab. Checked against five rival explanations before trusting any of it. Full method and sources further down this tab.
When. 2010 to 2023, latest year available per country. The 2030 outlook appears later in this tab.
Some of this is already caveated elsewhere on this page; some of it isn't stated anywhere else. Collected in one place, before any of the findings below:
FAO data is national per-capita food supply, not what women actually eat, and it cannot see who gets what inside a household.
The starch to anemia link is cross-country. The within-country and lagged tests found nothing, so this shows who's at risk today, not what would fix it tomorrow.
FAO balance sheets omit several of the worst: Mali 56.5%, Benin 50.0%, Chad 46.5%, Togo 41.5%, Burundi 39.0%. Every relationship here is likely a floor.
The World Bank anemia series is partly model-interpolated between surveys, which blunts year-to-year signal. DHS measured anemia covers 49 countries.
99 FBS item codes were mapped to 15 SUA food groups by the team, not by FAO. Every decision is documented in the pipeline doc.
Around nine variables were tried. Malaria and starch are large enough to be unlikely flukes, but they deserve held-out validation.
Countries are compared at whatever year is most recently available for them, not one shared year. A 2023 figure sits next to another country's 2015 figure.
The starch trap is the riskiest third of countries on imports, the riskiest third on starch, and the bottom third on micronutrients. Draw the cutoff elsewhere and the list changes.
Rival explanations were ruled out income first, then malaria, water, fertility and iron supplementation, in that order. A different order could shift how much credit each one gets.
Income enters as its logarithm, assuming an extra dollar matters more to a poor country than a rich one. And every effect is treated as a straight line that adds up on its own, with nothing amplifying or dampening anything else.
The question assumes two things: that half the world's calories now cross a border, and that the traveling half is the less nourishing one. Both are testable, and neither had been checked on this data. So that is where the analysis starts.
Across 178 countries we divided imported calories by total calories supplied, then averaged country by country, year by year. The share rose from 43.0% in 2010 to 50.4% in 2023, about +0.64 points a year, without a single reversal (r = 0.98). The average national diet crossed the halfway line in 2020.
Weighting by population gives the same direction, +0.40 points a year, while population-weighted starch share actually fell by 0.14 points a year. More trade has not meant starchier diets on its own. The damage is where the two overlap.
The second claim holds too, and for a mundane reason: the world ships what keeps. Sugar, oil and grain survive a month at sea. Leafy greens and eggs do not.
Cereals, oils and sugar are 60% of the average country's calories but only 37% of its iron. Vegetables are 2% of calories and deliver 13% of the iron and 26% of the vitamin A.
No FAO file labels a calorie as imported and then reports its nutrients, so that measure had to be built. Each food group's calories and nutrients get split by that group's own import share, the pieces are summed, and the two halves are compared per 1,000 kcal against a woman's daily requirement.
146 of 178 countries show a positive gap, median +33% of daily need per 1,000 kcal. Protein says the same: 26.7 g per 1,000 kcal imported against 32.9 g home-grown, thinner in 131 of 177 countries. By starch it is stronger still, with imported calories starchier in 84% of countries.
Import dependence does not predict anemia in women. Not across countries, not within a single country tracked over thirteen years, and not three years ahead of time. The correlation is −0.18, and the sign points the wrong way: countries buying the most food from abroad trend slightly healthier.
The most import-dependent economies on earth are Belgium and the Netherlands. The starchiest diet belongs to DR Congo, which imports 7.2% of its food and grows the rest. Starch and trade are negatively correlated globally. The countries eating the most starch are the ones trading least.
We are publishing this because it is the finding, and because a reader who can't see what we ruled out has no way to weigh what we kept.
We define the starch trap as a country that is simultaneously in the top third for imported calories, the top third for starch share, and the bottom third for micronutrient supply.
Each condition on its own points at the wrong countries. High imports alone catches Macao, the Netherlands and Belgium, who buy a diverse, low-starch diet from abroad. High starch alone catches DR Congo, 83% starch and 7% imported, which grows its own staples, so trade is not the route. Only the intersection isolates the countries the question is actually about.
| Country | Calories imported | Starch share | Micronutrient index | Anemia, women |
|---|
And the food it buys carries about a third of the protein density of what it grows.
A micronutrient index of 0.66: about half the iron women need, a third of the vitamin A, under a tenth of the B12.
The second-highest rate anywhere in the panel.
Tracing imports item by item for the heaviest-burden countries: more than nine tenths of what they buy is classed as starchy. Rice, wheat and maize, with palm oil and sugar alongside.
| Country | Group | Imports that are starchy | Leading starchy imports | Anemia |
|---|---|---|---|---|
| Liberia | Trap | 96.0% | Rice (100%) · palm oil (57%) | 42.5% |
| Afghanistan | Watch list | 95.9% | Wheat (46%) · palm oil (100%) | 45.4% |
| Haiti | Watch list | 93.7% | Rice (83%) · wheat (100%) | 45.9% |
| Gambia | Trap | 92.3% | Rice (100%) · sugar (59%) | 46.1% |
| Lesotho | Trap | 91.3% | Maize (88%) · wheat (100%) | 34.2% |
National food supply is an average, and women often do not receive the average. Iron needs peak during the reproductive years, nutrient-dense foods cost more, and women are more likely to go without them.
Niger's food supply delivers 24.2 mg of iron per person per day, above the commonly cited 18 mg target, yet 47% of its women are anemic. The reason is absorption, not supply: iron from meat, fish and eggs is absorbed efficiently, iron from cereals and beans largely is not, and only 4% of Niger's iron comes from animal sources. Across countries, total iron supply barely tracks anemia at r = −0.21, while the animal-source share tracks it 2.7 times more strongly at r = −0.55.
| Country | Iron supplied, mg/day | From animal sources | Anemia, women | GDP per person |
|---|
Guatemala is one fifth as rich as Australia and has lower anemia. Income is not destiny. Between 2010 and 2023 the typical country's iron supply rose from 15.4 to 16.0 mg a day, and rose in 102 of 162 countries. Anemia rose anyway in 73 of those 102. Burkina Faso, at 24.8 mg a day and 44% anemia, matches Niger's pattern.
Across three notebooks and fifteen scripts the team tried roughly nine explanations for why anemia sits where it does. Keeping the failures visible is the point: the ones that collapsed are what rules out the easy story.
Domestic calories are denser in 146 of 178 countries.
+0.315, still +0.235 with every control, and it survives the poverty test.
r = 0.525 with anemia. Adding it cuts the income coefficient from −2.85 to −0.35.
Animal-source share r = −0.55 against total iron r = −0.21.
r = −0.18 with anemia. Belgium and Bahrain both import over 96% of their calories; their anemia rates are 15% and 38%.
Country and year fixed effects across 2,276 country-years: no diet variable significant, all p > 0.13.
Lagged fixed-effects model, p = 0.22 to 0.72.
Local staple prices are falling in Lesotho and Mauritania. Iron-dense foods got cheaper relative to cereals in 9 of 10 checkable countries.
Total iron supply does not predict anemia once income is fixed (p = 0.75), and neither does exporting iron-rich harvests (p = 0.45).
With iron supplementation added, the 53-country interval just touches zero (p ≈ 0.07), so the effect size holds but the sample is small. The step-by-step version of this same test, with the exact number at each stage, follows below.
Some of the worst burdens sit outside the trap altogether. India has 53.7% anemia and imports only 6.8% of its calories, mostly edible oils, sugar and pulses, even though it is the world's largest vegetable-oil importer. Gabon has the worst rate in the panel at 59.9%, on $18,700 per person, with imports below the threshold. Their drivers are domestic diets, malaria, sanitation and health access, not trade.
Here's the same relationship walked through step by step. Starting point: starch share of diet correlates with anemia at +0.315 across countries, income already held fixed: a real, statistically significant relationship. Each row below adds one more rival explanation and shows what survives.
It shrinks. It never reaches zero. Five rival explanations, and the effect is still standing. Malaria is real and it absorbs 14% of the starch effect, leaving the rest intact. Handing out iron pills does not explain it away either.
If diet were a quick lever, a trap country cutting its starch should watch anemia fall. Two trap countries moved in opposite directions, and they show why the within-country test came up empty.
Starch share falling 1.04 points a year, the fastest diet improvement of any trap country.
Anemia rising 0.81 points a year anyway. 34.2% today, 39.4% forecast by 2030.
Starch share falling only 0.17 points a year, barely moving at all.
The only trap country where anemia is falling, down 0.68 points a year. 46.1% today, 41.9% by 2030.
Income is falling in several of the countries flagged here, most sharply Afghanistan, where GDP per person dropped by nearly a third. But growth is no escape either: Bhutan and Botswana both grew fast and are both still in the trap.
| Country | Group | GDP per person, 2010 | 2023 | Change |
|---|
Fitting each country's 2010 to 2023 line and extending it to 2030 named Haiti and Cabo Verde as new trap countries. Then the back-test ran: fit up to 2016, 2017 and so on, predict the years after, compare. The straight-line trend did worse than assuming nothing changes at all, 4.45 points of error against 3.73 on import share.
The model that survived damps the trend to 10% of its slope. Under it no country confidently crosses by 2030, and the uncertainty is about ±6.7 points on imports and ±4.4 on starch. So rather than name winners, the analysis measures how far each country sits from its last unmet condition, in cross-country standard deviations.
| Country | Distance to trap | Still missing | Conditions met | Observed drift |
|---|
"Closes in about N years" is the remaining gap divided by observed drift, where that drift is statistically significant. That is arithmetic on history, not a validated forecast. A wider lens finds 47 countries moving toward the trap on both fronts at once, including Samoa, Papua New Guinea, the Philippines and Namibia.
Globally starch is slowly leaving the plate, falling in 102 of 156 countries. In three trap countries and one watch-list country it is still climbing.
| Country | Group | Starch share, latest | Trend a year | 2030 on trend |
|---|---|---|---|---|
| Liberia | Trap | 72.0% | +0.65 | 77.0% |
| Eswatini | Trap | 56.9% | +0.46 | 59.1% |
| Haiti | Watch list | 57.9% | +0.42 | 58.5% |
| Mauritania | Trap | 52.6% | +0.29 | 56.2% |
Four forecasting models were tested on held-out 2019 to 2023 data. A per-country trend anchored at the last observation won with 1.14 points of error, beating naive persistence (1.45), fitted-line extrapolation (1.37), a random forest (1.49 to 2.22) and cross-sectional regression (about 8.0).
| Country | Anemia, latest | Trend/yr | 2030 | Why it matters |
|---|---|---|---|---|
| Mauritius | 32.8% | +1.01 | 39.3% | Fastest rise in the panel, 87% import dependent |
| Lesotho | 34.2% | +0.81 | 39.4% | In the trap and rising fast |
| Afghanistan | 45.4% | +0.81 | 50.6% | Watch list, imports climbing toward the line |
| Haiti | 45.9% | +0.29 | 47.8% | Watch list, third closest to the trap |
| Mauritania | 54.5% | +0.10 | 55.2% | Highest anemia of any trap country |
| India | 53.7% | +0.29 | 55.6% | Not trade-driven. One lens is never enough |
WHO bands: 20 to 39.9% is moderate, 40% or more is severe. Twenty countries sit at the severe level today and eighteen are projected to in 2030, with only Myanmar newly crossing in. 118 of 156 countries still have a rising trend.
Haiti already meets the starch and micronutrient conditions and needs about three more import points. Cabo Verde meets the import and micronutrient conditions and hovers 1.3 points under the starch line. The straight-line fit sends both across; the damped model that survived the back-test does not.
The red line is the trap threshold: 63.7% for imports, 50.2% for starch. The band is the back-test error, about ±6.7 points on imports and ±4.4 on starch by 2030, which is why neither country counts as a confident crossing.
Whatever world prices do, local markets are calmer right now. Staple prices are falling in six of the eight trap and watch-list countries with current readings, which is part of why food prices failed as the channel.
| Country | Group | Change against a year earlier |
|---|
Average year-on-year change across each country's tracked staples, readings from June 2026. Too old to count as current: El Salvador +72.5% (Aug 2022), Lebanon +263% (Feb 2023), Liberia +11% (Oct 2020), Haiti −6.5% (Feb 2025).
One policy does not fit all. Sorting countries by why they are at risk is what decides which action reaches them.
Trap countries buy most of their grain abroad, so world prices reach them through the import bill. If the last decade repeats, wheat rises 17 to 57% by 2030 depending on grade and method, maize 20 to 41%, rice 14 to 19%. Chicken is the exception, falling about 14%, which would make one well-absorbed source of iron cheaper.
Global export prices in US dollars, not adjusted for inflation, projected from their 2023 to 2025 average. These show where prices go if the last decade continues; they are not forecasts of what any country will pay. Local staple prices are calmer today, falling in six of the eight trap and watch-list countries with current data, and prices showed no measurable link to anemia. Read this as budget pressure on import-dependent governments, not as a cause.
Make iron, folic acid and vitamin A fortification a condition of import for wheat flour, rice, maize and edible oil, which are the exact calories trap countries import most. The imported half then delivers micronutrients instead of diluting them.
Supply is concentrated enough for this to work: the top three exporters ship 68 to 78% of traded rice, 61 to 66% of maize and 83 to 86% of palm oil. First movers would be Gambia and Liberia, where rice is about 100% imported, Lesotho on maize and wheat, and Djibouti on nearly everything.
Cut tariffs and VAT on pulses, dried fish, eggs and milk powder, and use state procurement and food-aid contracts to buy them rather than only grain. Spread purchases across suppliers so one shock cannot empty the shelf.
Chicken is the one iron-rich import projected to get cheaper, about 14% by 2030. Watch concentration risk: 78% of traded lentils come from three exporters.
Fund home and school gardens, cold chains and extension services for vegetables, pulses, small livestock and eggs, and screen trade policy so cheap imports do not crowd out local crops.
For Morocco, Cabo Verde and Haiti, this is the way to hold the last unmet threshold.
National averages hide who goes without. Reach women directly with weekly iron and folic acid supplementation through antenatal care, schools and workplaces, and add cash or vouchers earmarked for eggs, fish, meat and pulses.
Measure intake by sex, not by household.
The three conditions can be recomputed annually from the Food Balance Sheets FAO already publishes. Flag any country within 0.25 standard deviations of its last threshold for a nutrition review: today that is Morocco, Cabo Verde, Haiti and El Salvador.
Pair the screen with the anchored anemia forecast, and confirm the timing with anemia data itself.
Where anemia is high but trade is not the route, pair nutrition work with malaria control and with water and sanitation.
Split anemia by cause with IHME GBD data, pull a malaria time series rather than a single year, and study fortification mandates as a natural experiment.
The core assumptions behind every number on this page are stated up front, in "Why this, what we're asking." What follows here is the remaining chart-specific fine print.
It reflects which food groups get traded, not whether imported wheat is worse than local wheat. Vitamins A and C dominate it.
2021 only, and missing for five trap countries, so it explains differences between countries and not change over time.
World US-dollar prices, not adjusted for inflation, and not what any country actually pays.
Every FAOSTAT file has the same shape: country, item, element, year, value. One row is one fact. We joined five sources on country and year, then ran the same test repeatedly while adding rival explanations one at a time.
anemia% = 40.02 + 5.07 × import share + 31.98 × starch share + 0.49 × micronutrient adequacy − 3.15 × log(GDP per capita). Fitted on 167 countries. R² = 0.463. The model explains 46% of the variation; the rest is healthcare, genetics, fortification and geography. South Asia alone carries 11 extra points that income and diet together can't account for. The starch coefficient has z = 5.435 and p < 0.001, with a confidence interval of 20.45 to 43.52, nowhere near zero.
Income correlates with starch share at −0.759. Richer countries eat less starch and have less anemia, so any raw starch-anemia relationship could simply be poverty wearing a costume. Holding income constant was the first thing we did, not the last, before adding malaria, water, fertility or iron supplementation.
Comparing different countries risks confusing "starchy" with "poor" or "African" or "rural". Panel fixed effects compare each country only against itself, cancelling every permanent trait. Nothing survived that test. The honest reading is that the relationship is structural rather than a lever: it tells you who is at risk today, not that changing diet tomorrow would fix it.
| Dataset | Used for | Coverage |
|---|---|---|
| FAOSTAT Food Balance Sheets | Imports, production and calories by item, giving import share and starch share | 4.82M rows · 178 countries · 2010 to 2023 |
| FAOSTAT Supply Utilization Accounts | Iron, zinc, vitamins A, B12 and C, and protein by food group | 743K rows · 179 countries |
| World Bank WDI, anemia | The outcome variable, women 15 to 49 (SH.ANM.ALLW.ZS) | 2010 to 2023 |
| World Bank WDI, GDP per capita PPP | Income control (NY.GDP.PCAP.PP.KD) | 2010 to 2023 |
| World Bank and WHO GHO | Malaria, water, sanitation, fertility, stunting, anemia uncertainty bounds | 195 countries |
| DHS Program | Measured anemia and iron supplementation for the poverty test | 49 to 59 countries |
| FAOSTAT Food Security Indicators and Cost of a Healthy Diet | Gendered food insecurity and food-group costs | 118 to 135 countries |
| FAOSTAT Detailed Trade Matrix | Exporter concentration and supplier dependency | Bilateral flows |
| WFP Global Market Monitor and World Bank Pink Sheet | Local and global food-price checks | 81 countries · 1960 to 2026 |
| File | What it produces |
|---|---|
| analysis/scripts/00–14_*.py | Crosswalk, features, density gap, classification, forecasts, validation, proximity, confounders |
| analysis/trade_and_nutrition_analysis.ipynb | End-to-end narrative notebook, steps 1 to 6 |
| country_snapshot_classified.csv | Trap classification, scatter, correlations |
| nutrient_density_gap_latest.csv | Imported against domestic nutrient labels, density histogram |
| global_import_share_trend.csv · global_trend_stats.txt | The import-share trend |
| regression_cross_section_2023.txt · regression_with_confounders.txt | The starch effect and its controls |
| regression_panel_fe_summary.csv · regression_early_warning_lag3_summary.csv | Within-country and early-warning tests |
| forecast_2030*.csv · forecast_validation.txt · trap_proximity*.csv | The 2030 outlook and watch list |
| master_panel*.csv · data/external/* | Country-year panel, WHO and DHS merges, coverage reports |
| malaria_incidence_2021 · iron_export_anemia_country_year · anemia_trend_forecast_2030 | Malaria split, iron-supply trend, anemia outlook |
| iron_food_price_forecast_2030 · wfp_country_iron_food_price_signal | Price outlook and local price signal |
More imports, starchier diet, more anemia: that chain is the intuition this project started from, and the matrix is where it visibly breaks. Import share is negatively linked to starch share and barely linked to anemia at all. Starch share is the variable that tracks anemia, though it is tangled with income.
Pearson r, 2023 snapshot, 178 countries. Warm is positive, cool is negative, and intensity is strength. Sources and method notes sit in "Method, sources & limits," above.