Editorial standard · applies to every event page
How DeathVault counts the dead
A death toll is never a single number that someone found. It is an estimate, built with a method, from sources of unequal quality. This page explains the method behind every figure on this site — so you can judge each one, cite it responsibly, and disagree with it precisely.
Short answer
Every event on DeathVault carries three numbers: a low estimate, a high estimate and a central figure. The low and high come from the most credible published scholarship at each end; the central figure is our reading of where the weight of evidence sits — never the most dramatic number in circulation. Each page lists the sources used and, where the field is divided, a table of estimates by source.
What the archive covers
36
Events
69
Cited references
36/36
With low–high range
36/36
With on-page sources
1. Where the numbers come from
We rank sources in a fixed order and use the highest tier available for each event. Tier 1: institutional datasets with a published method — WHO and UN agencies, national statistical offices, the IAEA, the Uppsala Conflict Data Program, the Correlates of War project. Tier 2: peer-reviewed historical demography and academic monographs (for example Rummel, Clodfelter, White, McEvedy & Jones). Tier 3: reference works with editorial review — Britannica, the Holocaust Memorial Museum, national archives. We do not use news reports, encyclopedic wikis or advocacy sites as a primary source for a figure, although we may cite them for context.
For each event we record every source consulted and surface it on the event page under References. When two Tier 1 or Tier 2 sources disagree by more than 25 %, the page shows an estimate-by-source table rather than a single figure.
2. Ranges, central figures and why they differ by millions
For anything before the twentieth century the surviving records are incomplete, and even modern conflicts are counted with different definitions. The Second World War is documented as 70 to 85 million dead not because historians are careless but because they disagree about whether to count famine deaths in Bengal and China, disease among refugees, and the Soviet civilian toll — each of which moves the total by millions. A range is the honest answer; a single number would be false precision.
Our central figure is the median of credible estimates when the distribution is symmetric, or the figure adopted by the best-documented source when it is not. We never average a serious estimate with a fringe one, and we never pick the highest figure to make an event look larger.
3. Direct deaths versus excess mortality
There are two legitimate ways to count. Direct deaths are people killed by the event itself — a bomb, a battle, a diagnosed infection. Excess mortality is how many more people died during the event than would have died without it, which captures famine, disrupted healthcare, displacement and untested disease. Excess mortality is almost always larger, sometimes several times larger. COVID-19 is the clearest modern example: about 7 million confirmed deaths against 18 to 28 million excess deaths.
Each event page states which measure the central figure uses. When both exist we show both. When only direct deaths are documented — typical of ancient and medieval events — we say so, and the range is wider to reflect what is missing.
4. Wars: soldiers, civilians and the deaths nobody records
Military deaths are the best-documented deaths in history: armies keep rolls. Civilian deaths in war are the worst-documented, because the institutions that would count them have often been destroyed. For every war we try to separate the two, and we include war-attributable famine and disease when a credible demographic study exists for them (as it does for the Thirty Years' War, the Taiping Rebellion and the Second Congo War). We do not include deaths from a conflict's long-term economic aftermath.
5. Famines and genocides: counting absence
Famine deaths are estimated as a demographic deficit: the gap between the population a census expected and the population it found, corrected for births that did not happen and for migration. That is why famine figures have the widest ranges on this site — the Great Chinese Famine is documented at anywhere between 15 and 55 million. Genocide figures depend on legal and historical definitions of who was targeted; we follow the definition used by the tribunal or commission that investigated the event, and we say so on the page.
6. Pandemics: confirmed, estimated and modelled
For modern outbreaks we distinguish laboratory-confirmed deaths (what a ministry of health reports), estimated deaths (confirmed plus a correction for testing gaps) and modelled excess deaths. For historical pandemics — the Black Death, the 1918 influenza — only demographic reconstruction exists, so we cite the range of the reconstructions and the assumptions behind them. Pandemic pages also live on our sister site PlagueAtlas, which is the canonical home for that topic.
7. Nuclear events: acute versus lifetime
Radiation kills in two very different time frames. Acute deaths — the blast, the fire, radiation sickness in the following weeks — are counted precisely. Lifetime deaths from radiation-induced cancer are statistical projections with enormous uncertainty: for Chernobyl, from about 4,000 (WHO/IAEA) to tens of thousands in other models. We show acute deaths as the central figure and the projections as the range, with the source of each.
8. Ongoing events and corrections
Events still in progress — active wars, current outbreaks — are updated as institutional data is published, and every page carries a last-updated date. We correct mistakes quickly: if a figure on this site does not match its cited source, or a better source exists, write to us through the Contact page. Corrections from historians and demographers are especially welcome, and we credit them.
9. How to cite DeathVault
Cite the event page, its central figure and its range, and the underlying source we list. DeathVault is a secondary source: for academic work, follow the reference through to the primary study. Data on this site is published under a Creative Commons Attribution 4.0 licence; you may reuse it with a link back.
Frequently asked questions
Why do different websites give different death tolls for the same event?
Because they choose different definitions (direct deaths or excess mortality, military only or civilians too), different sources, and different points inside the credible range. DeathVault publishes the range, the central figure and the source, so the choice is visible instead of hidden.
Is the central figure an average?
Not a simple one. It is the median of credible estimates when they cluster, or the best-documented estimate when they do not. Fringe figures — extreme highs used for advocacy or extreme lows used for denial — are excluded from the calculation and, when notable, mentioned in the text.
Are the totals on the home page a real number?
They are the sum of the central figures of every event in the archive. Because events overlap (the Holocaust is part of the Second World War's civilian toll; the 1918 influenza overlaps the First World War), the total is an index of what the archive documents, not a count of distinct people.
What counts as a mass-death event for inclusion?
A discrete event, with a start date, in which a credible estimate places the death toll above 100,000, and for which at least one Tier 1 or Tier 2 source exists. Long-running causes of death without a start — endemic malaria before 1900, road accidents — are not events and are not included.
Who writes and checks the pages?
DeathVault is written and maintained by Furiosa Studio, an independent studio. Every event is cross-checked against at least one institutional dataset and one independent academic or reference source before publication, and the citations are listed on the page.