Conflict researcher Paola Vesco and her team have developed a model that helps measure the uncertainty around such conflict death estimates. Their approach takes into account the tendency of conflict datasets to underestimate fatalities, and provides a more realistic range of possible death tolls. They explain how their model could improve estimates in conflicts like Sudan’s.

What do we know about the number of dead in Sudan?

The number of people killed in Sudan’s war since it began in April 2023 remains contested. Deaths in 2025 have been especially hard to establish as communications were cut and humanitarian access was blocked.

In 2025, fighting intensified as the Rapid Support Forces – a paramilitary group which has fought the Sudanese army for control of the state since 2023 – advanced in Darfur and North Kordofan. Civilian deaths and displacement rose.

In February 2026, a UN fact-finding mission concluded that the Rapid Support Forces’ killings in El-Fasher bore the hallmarks of genocide.

The group’s capture of El-Fasher in the Darfur region in October 2025, after an 18-month siege, marked a significant territorial and strategic shift in the war. It consolidated the Rapid Support Forces’ control over much of Darfur and strengthened their position in western Sudan. It was also one of the war’s deadliest episodes.

ACLED, a global conflict monitoring organisation, reported around 19 000 deaths, which it describes as a conservative minimum count of deaths directly caused by political violence in Sudan in 2025.

The Uppsala Conflict Data Program, a research team at Uppsala University that has recorded conflict deaths worldwide for decades, estimated that attacks against civilians killed at least 60 000 people. More than 10 000 died in fighting between the army and armed groups.

Drawing on survivors’ reports, satellite imagery and other evidence, Yale’s Humanitarian Research Lab estimates that at least 60 000 people were killed during and after the paramilitary group’s capture of El-Fasher between late October and the end of December 2025.

Even worse, these figures capture only direct deaths from violence in Sudan. Many more have probably died from disease and starvation, as the Rapid Support Forces built earthen walls around El-Fasher in an effort to starve out the population.

Why are the estimates so different?

The true number of war deaths can never be known with certainty, but scholars understand reasonably well why estimates vary.

Much of the difference across reports stems from how organisations select and handle sources.

The Uppsala Conflict Data Program’s estimate of 60 000 killed in El-Fasher draws directly on Yale’s Humanitarian Research Lab. It is corroborated by independent reporting from Sudanese media, NGOs and UN agencies.

Beyond this “best” (most likely) figure, Uppsala’s programme also keeps a “high” estimate (75 000), reflecting the upper end of Yale’s assessment, and a markedly lower “low” estimate to flag that the exact number of deaths is uncertain.

ACLED, on the other hand, reports a single fatality number. When sources disagree, it takes the most conservative figure.

The differences in the estimates of deaths also stem from how monitoring organisations process information.

Both the Uppsala Conflict Data Program and ACLED record the exact date and location of violent incidents when these are available. But ACLED only counts deaths that can be linked to a specific incident. Deaths reported only as a broad total, without identifiable incidents, may therefore be left out. Uppsala’s programme instead records them as a “summary” event, which presents the total death toll over a longer period or a wider area, rather than for a single attack.

This apparently simple rule matters enormously when many smaller, undated incidents add up to a large total, as in El-Fasher.

What did you do differently and what numbers did you come up with?

Providing an uncertainty estimate – a plausible range rather than a single number of people who may have died in war – reflects how much confidence an estimate deserves, and invites caution.

Still, most sources report only one number.

In a recently published article, we built an uncertainty estimate. We surveyed Uppsala Conflict Data Program’s expert coders to build a model converting any single number into a range of people that have plausibly died in a given war. This range is known as a “probability distribution”.

We applied this model to Sudan in 2025. The results are provided in the figure below. For each number of deaths on the horizontal axis, the height of the curve shows how plausible that number is according to our model. The higher the curve, the more plausible the number.

The peak of the curve from our model indicates that the most likely death toll in Sudan’s war in 2025 is close to 80 000 deaths. As our model is estimated from the Uppsala programme, it is unsurprising that this estimate is close to what they report.

We find that the true number of people killed in 2025 is very likely (with a 90% probability) to fall between 73 000 and 90 000, and there is a small chance that the actual number is as high as 100 000.

This relatively small range indicates that the uncertainty around Sudan’s deaths is much narrower than assumed – even the lower bound number far exceeds many official estimates.

Our estimates, however, should be read with caution. They are informed by the Uppsala Conflict Data Program’s knowledge and only indirectly reflect other estimates. Crucially, our model results cannot be checked against the true number of deaths, which tends to remain unknown as conflict actors have incentives to conceal it.

Despite these caveats, providing a range of people that plausibly have died in war instead of a single figure gives weight to the estimate, and leaves little room to dismiss inconvenient numbers as unreliable.

Why are these data points important?

Understanding how conflict death estimates vary across sources, and why, increases transparency.

Around the attack on El-Fasher, the uncertainty around death reports became a political issue. We argue that it should be a scientific one. One that can be measured.

The evidence suggests that the El-Fasher massacre could have been prevented. Throughout the siege, NGOs and academic institutions repeatedly warned foreign officials of an impending bloodbath.

Pressure against the United Arab Emirates – the Rapid Support Forces’ main supporter – could have curbed the flow of weapons to the group. Yet, western politicians prioritised economic and political ties with Abu Dhabi over preventing atrocity. Officials looked the other way, ignoring or downplaying the reports of rising deaths as uncertain.

Since fighting broke out between the army and the Rapid Support Forces, Sudan has become one of the world’s worst humanitarian and displacement crises. More than 34-million people – over half the country’s population – need humanitarian assistance, including food aid. An estimated 14-million people have been displaced.

This suffering could have been foreseen and minimised. Even in the most conservative case, our model finds that more than 70 000 people died in 2025, most of them within a few months. Are 70 000 deaths enough to look the other way?

Conflict monitoring organisations such as the Uppsala Conflict Data Program have documented war deaths for decades, drawing on vetted sources and consistent coding rules. Since the uncertainty around their figures can be quantified, as our model shows, this uncertainty is no longer a reason for delay or inaction.

Decision-makers often have the information they need to act early. Used in time, this information can help prevent atrocities and save lives.

Written by Paola Vesco, Senior Researcher, Peace Research Institute Oslo (PRIO); Amber Deniz, Senior Analyst, Uppsala University; David Randahl, Associate Senior Lecturer in War Studies, Swedish Defence University; Johan Brosché, Associate Professor, Department of Peace and Conflict Research, Uppsala University, and Magnus Öberg, Director of the Uppsala Conflict Data Program, Uppsala University

This article is republished from The Conversation under a Creative Commons license. Read the original article.