Methods

How the Gross Domestic Suffering index is built

Every data source, formula and adjustment behind the index, so anyone can check our work, challenge it, or rebuild it from the open code.

In brief

We score 12 indicators of suffering for every country and every year from 2000 to 2023, scale each from 0 (least suffering) to 100 (most) against fixed goalposts, average them into six equally weighted pillars, and average the pillars into one score. Every indicator counts suffering, so higher is always worse.

What we measure

The index measures the amount of suffering in people's lives, and in animals' lives, rather than how long lives last. Early death enters for a different reason: it causes some of the deepest suffering there is for the people left behind, from parents burying children to children losing a parent.

PillarIndicatorWhat it countsSource
IllnessPremature adult deathProbability of dying between ages 15 and 60, per 1,000WHO
Infant mortalityDeaths before age 1 per 1,000 live birthsWorld Bank / UN IGME
Share of life lived in ill health(Life expectancy − healthy life expectancy) ÷ life expectancyWHO
Alcohol and drug deathsDeaths from alcohol and drug use disorders, per 100,000WHO Global Health Estimates
Gender and sexual discriminationGender Inequality Index (85%)Women's losses in reproductive health, empowerment and work (0–1)UNDP
Anti-LGBTI laws (15%)Laws that criminalize or exclude LGBTI people, net of laws that protect themVelasco, via Our World in Data
ViolenceViolent deathsHomicides, deaths in armed conflict and suicides, per 100,000UNODC, UCDP, WHO
UnfreedomPeople held captivePrisoners plus people in modern slavery, per 100,000World Prison Brief, Walk Free
OverworkShare of workers working 49 or more hours a weekILO
DeprivationUndernourishmentShare of people without enough foodFAO
Extreme povertyShare living on under $3.00 a day (2021 PPP)World Bank
Forced displacementRefugees, asylum seekers and internally displaced people from the country, per 100,000UNHCR
Animal sufferingFarm-animal mistreatmentGrade for laws protecting farm animals, A (strongest) = 0 to G (weakest) = 100World Animal Protection

The formula

  1. Orient. Every indicator is framed so that a higher value means more suffering.
  2. Transform. Rates that span orders of magnitude (deaths, prisoners, displaced people) are put on a log scale, so a change from 1 to 10 counts as much as a change from 10 to 100.
  3. Scale with fixed goalposts. Each indicator is scaled from 0 to 100 between the 2.5th and 97.5th percentiles of all observed country-years, 2000–2023, and capped at 0 and 100. The farm-animal grade is already on a 0–100 scale. The goalposts never move between years, so a country's score changes only when its own data changes. This is what lets us compare a country with its own past, as well as with other countries.
  4. Average into pillars. Each pillar is the average of its indicator scores. The one exception is Gender and sexual discrimination, weighted by the people each indicator covers: women are about half of people, and LGBTQ+ adults about 9% (Ipsos, 2023), so the Gender Inequality Index counts 85% and anti-LGBTI laws 15%.
  5. Average into one score. The GDS score is the average of the six pillars, each weighted equally.

A country is scored only when it has data for at least five of the six pillars, including Illness and Animal suffering.

Uncertainty

Equal weights are a judgment, not a fact. To show how much each rank depends on them, we recompute every country's rank under 1,000 random sets of pillar weights and publish the range covering 90% of results. A narrow range means the rank holds whatever weights you choose.

Deliberate choices

What we leave out, and why

Adjustments to source data

We use public sources as published, with the following exceptions. Each one is also documented in the code.

  1. Japan's overwork figure comes from Japan's own Labour Force Survey. The ILO publishes no breakdown of working hours for Japan. We use the share of workers working 49+ hours a week from Japan's Labour Force Survey, as published by the Japan Institute for Labour Policy and Training (JILPT): 23.1% in 2010, falling to 15.2% in 2023. JILPT's figures for other countries match the ILO's, so the two sources are comparable. China is also missing from the ILO data and, for now, has no overwork figure.
  2. Modern slavery is a single estimate, applied to every year. Walk Free's Global Slavery Index 2023 is the only edition with comparable country estimates, so the same figure is used for 2000–2023. Changes in the Unfreedom pillar over time therefore come from prison populations and working hours, not slavery.
  3. Missing slavery estimates are filled from comparable neighbours. Walk Free does not estimate some small countries. For small Western European states we use the median of close neighbours: Iceland, the Faroe Islands and Greenland get the Nordic median; Luxembourg gets the median of Belgium, the Netherlands, Germany and France; Liechtenstein, Monaco, Andorra, San Marino, Malta, Gibraltar, the Channel Islands and the Isle of Man get their nearest neighbours. Other missing countries get the median of countries in the same World Bank region and income group. These countries are flagged as imputed.
  4. Treatment of farm animals comes from 50 countries, with regional medians for the rest. World Animal Protection's Animal Protection Index (2020) grades the farm-animal laws of 50 countries from A (strongest) to G (weakest). Those countries eat 88% of the world's land animals. Every other country gets the median grade of rated countries in its World Bank region, flagged as imputed. The 2020 grades apply to every year, so this pillar does not change a country's trend over time. The Netherlands profile has no farm-animal section, so we use its overall grade, B.
  5. The LGBTI index never stands in for missing gender data. It carries only 15% of its pillar, so where the Gender Inequality Index is missing the whole pillar counts as missing.
  6. Missing suicide rates are filled from comparable countries. The WHO publishes no suicide rate for some very small states. They get the median of countries in the same World Bank region and income group, flagged as imputed.
  7. Alcohol and drug deaths end in 2021. The WHO Global Health Estimates run from 2000 to 2021, so 2022 and 2023 carry the 2021 figure forward, flagged as imputed.
  8. Homicide data uses WHO estimates where UN data is missing. Our first source is the UN Office on Drugs and Crime. Where it has no usable figure, we use the WHO's modelled homicide rate, flagged as imputed.
  9. No record means zero for war deaths and displacement. The Uppsala Conflict Data Program records every armed conflict, so a country with no entry has no recorded conflict deaths. Likewise, a country UNHCR never lists as a place of origin has no recorded refugees or displaced people. War deaths are counted in the country where they occur.
  10. Gaps between survey years are filled within each country. We interpolate linearly between observed years, and carry the first or last observation flat for up to 5 years (10 years for homicide, prison, poverty and working-hours data, which are published irregularly). We never borrow another country's data to fill a time gap. Every filled value is flagged, and each country-year reports its share of imputed values.
  11. Very low hunger is reported as 2.5%. The FAO does not estimate undernourishment below 2.5%, so all countries under that threshold share the same best score.

Farm-animal grades

Farm-animal gradeABCDEFG
Animal suffering score01733506783100

Known limitations

Open data and code

The full dataset (every country, every year, every indicator, raw and scaled, with imputation flags), the fixed goalposts and the code that builds them are on GitHub. We welcome corrections and better data.

Last updated 2026-11-19.