Clay Tech

"clay-works make things real"

translated from clazytech.com

Ten Thousand Years of Violence Data Say We Live in Peaceful Times

original Society & Ideas

“We live in a time of much conflict, a hard time to survive.”

I’ve been hearing that phrase a lot lately. In the news, in casual greetings.

Every time I hear it, something nags at me a little.

Is it actually true? Has anyone counted?

Fortunately, someone had. The tools were all there, from bones to databases. So I lined them up.

First, look at the graph

Let me fix the unit of measurement: violent deaths per year per 100,000 people. This includes both war and homicide. The sources differ by era, from traces on bones to statistical databases, but I’ve standardized everything to this unit.

The vertical axis is logarithmic. I’ll explain why later.

Long-term trend in deaths from violence and war (estimates)

1 10 100 1000 Deaths per 100,000 people per year (log scale)

The two World Wars

Recent slight rise

Prehistoric 3000 BCE Ancient Medieval Early modern 19th century 1900-45 Postwar 2000s 2020s

Central estimate

Range of estimates (disagreement across studies) Sources are listed in the references section at the end. Prehistoric through early modern figures are converted estimates drawn from multiple studies.

The line is the central estimate, and the band is the range of disagreement across studies.

Three things are worth noticing here.

The line falls as it moves rightward. From roughly 150 in prehistoric times to roughly 8 today. Down to about one-twentieth.

But it isn’t monotonic. There’s a massive spike from 1900 to 1945.

And the band narrows as it moves rightward. That’s because the quality of the historical record improves. The fact that the prehistoric band spans a full order of magnitude isn’t a fudge — it’s information, telling us that scholars genuinely disagree by that much.

This is why the vertical axis is on a log scale. On a linear axis, you’d face a choice: either the prehistoric bar swallows everything, or the modern-era changes disappear. Fitting both a twentyfold decline and a 20th-century spike into the same picture requires a log scale.

What I referenced, and how I counted

The method of counting differs completely by era. Let me be upfront about that.

Prehistory (before state formation). Bones are all we have to go on. Skull fractures, stab wounds, arrowheads still lodged in bone. We count what percentage of excavated human remains show signs of violent death. This field goes back to Keeley’s War Before Civilization (1996), which estimates an average of around 15% violent death rates in non-state societies, while Fry and colleagues counter that the rate is much lower if you restrict to mobile hunter-gatherers. Nakao and colleagues’ comprehensive study of Jomon-period skeletal remains in Japan (2016) put the figure at 1.82%. Even within “prehistoric,” the numbers differ by an order of magnitude. That’s why the band is so wide.

To convert “percentage of deaths” into “per 100,000 people per year,” you multiply by the crude death rate of the time (assumed at 30-40 per 1,000 people per year). At the Jomon rate of 1.82%, that comes to roughly 60 per 100,000 per year; at Keeley’s 15%, roughly 500.

Ancient through medieval. For the Middle East, a 2023 study reconstructed violence trends from 12,000 BCE to 400 BCE using trauma data from 3,500 excavated skeletons (Baten et al., Nature Human Behaviour). There’s a peak in the Chalcolithic period (roughly 4500-3300 BCE), when cities emerged, followed by a steady decline over the next two millennia. The authors read this as running parallel to the emergence of legal systems, centralized administration, and religious institutions.

Medieval through early modern Europe. From here on, we have written records. Eisner’s long-term series of homicide rates, compiled from court records and similar sources, is the definitive one: it falls from 30-50 per 100,000 per year around 1300 to roughly 1 by around 1900.

The 20th century onward. This is the age of statistics. Battle deaths are counted by UCDP (the Uppsala Conflict Data Program), homicides by UNODC. During the two World Wars, the global average reached 200-300 per 100,000 per year, exceeding even the prehistoric central estimate — the only reversal on the graph. In the postwar period the rate has stayed almost consistently low, with battle deaths in the mid-2000s at around 0.5 per 100,000 per year. Probably the lowest figure in human history.

There were four turning points

Each bend in the graph corresponds to an event.

First, the emergence of the state (from roughly 3300 BCE). The decline from the Chalcolithic peak into the Bronze Age coincides with the period when courts and administration began replacing revenge. It’s close to an archaeological confirmation of what Hobbes argued in theory in Leviathan.

Second, the monopolization of criminal justice by the early modern state (roughly 1500-1900). Europe’s homicide rate fell from 30 to 1 — down to a thirtieth. Dueling was outlawed, and private retaliation gave way to courts. This is what Elias called “the civilizing process.”

Third, the early 20th century (1914-45). Industrialized total war. Having monopolized violence, the state also gained the capacity to exercise that monopolized violence with maximum efficiency. Roughly 3.1 million people died in Japan alone. These eight years, more than any phase of the Warring States period, account for the deadliest stretch in this country’s history, by an order of magnitude.

Fourth, the long peace (1945-). Nuclear deterrence, trade interdependence, international institutions. How much credit each factor deserves is debated, but the result — no war between great powers for 80 years — stands as a number.

And there’s a bend at the far right edge that may or may not turn out to be a fifth turning point. Since the 2010s, battle deaths in Syria, Ukraine, and the Middle East have risen to several times their historic low. The level is still near the historical minimum, but the direction is upward.

Discussion: the gut feeling is correct, but the sense of the era is mistaken

Here’s the main point.

“There’s more conflict now” is true compared to around 2005.

“We live in a time of much conflict” is false compared to 10,000 years.

These two propositions aren’t in conflict. They just use different filters. The former looks at the world through a window of the last 20 years; the latter through a window of 10,000 years. Both windows are legitimate in their own right.

Here’s something worth noticing.

People born from the latter half of the 20th century onward came of age near the lowest point in human history.

Someone born in 1970 has an observation window that includes the end of the Cold War and the trough of the 2000s. Measured from that trough, everything since looks like nothing but a rise. The gut feeling that “conflict is increasing” is correct, even in the data.

A generation that happened to grow up in the best valley in history looks at the view slightly up the slope from that valley and calls it “an age of mountains.” That’s essentially the structure of it.

There’s another issue: visibility. Footage from a battlefield 10,000 km away arrives on your phone every morning. This has never happened before in human history. Our grandparents’ generation lived through more wars in their own time, but escaped seeing most of them. The total volume of conflict has decreased, while the visible volume of conflict has increased. What we feel is probably proportional to the latter.

Finally, a dose of cold water.

You cannot conclude “so it will keep declining” from this graph. War death distributions follow a well-known extreme fat-tailed pattern, which makes it statistically dangerous to infer a trend reversal from a few decades of peace (Cirillo and Taleb). Japan itself has a precedent for this. After 250 years of Tokugawa peace came the worst eight years in its history. Nowhere on this graph is there a guarantee that the right edge will keep falling.

So let the conclusion stay modest.

Only three things can be stated with confidence. The present is, within what’s measurable, the least murderous era on record. The early 20th century was an exceptional spike. And a track record of peace is no guarantee of the future.

Even so, before saying “we live in a time of much conflict, a hard time to survive,” it’s worth counting first.

References

Original concept and direction: Kuzuryu. Writing: AI.

Originally published in Japanese at https://clazytech.com/2026/08/1755/. Translated with LLM assistance and reviewed before publication.