Counting the Dead

What Is a Relative Risk Ratio?

A Relative Risk Ratio (RR) compares the rate at which two groups experience an outcome - in this case, death in armed conflict. The concept comes from epidemiology [1][2]. Guha-Sapir & van Panhuis applied it directly to conflict mortality in a 2003 Lancet paper [3], and Ayoub et al. (2024) developed a closely related “Index of Killing Civilians” to compare Gaza operations over time [4].

• RR >> 1 - Military personnel die at a much higher rate than civilians. Expected when combatants engage each other and civilians are largely spared.
• RR ≈ 1 - Civilians die at roughly the same rate as military personnel. The distinction between combatant and civilian has collapsed.
• RR < 1 - Civilians face higher proportional risk than the military. Characteristic of genocides and mass atrocities.

The Full Spectrum

All 44 conflicts on a compressed logarithmic scale, sorted by RR. Where sources disagree, the bar is split into segments - each segment represents a different source’s estimate. Hover any bar to see source details. Click any bar to jump to its data.

Every legally recognized genocide - Rwanda, Srebrenica, Cambodia - clusters at or below RR = 1. The atomic bombings and Nanjing sit in the same zone.

Nagasaki (0.18) has the lowest RR on the chart. Unlike Hiroshima - which housed the Second General Army headquarters - Nagasaki’s bomb detonated over the Urakami Valley, an industrial district and its residential population. The military garrison of ~9,000 was largely outside the blast zone; of ~74,000 dead, only ~150 were soldiers killed instantly [35].

Hiroshima’s RR is itself disputed (0.59–1.5) depending on which garrison estimate you use. With ~14,000 military personnel in the blast zone, RR ≈ 0.59. But if the full Second Army garrison of ~43,000 is included, more military deaths push the RR up to ~1.5 - crossing the threshold from “mass atrocity” into “indiscriminate.”

Urban Combat Close-Up

Urban warfare is where the RR question gets hardest. This chart compares military and civilian death rates across nine urban battles. Click any bar to jump to its data.

Grozny II is the standout - the only urban battle where the civilian death rate (14.4%) exceeded the military rate (5.0%). Russia flattened the city while a mostly elderly civilian population remained trapped. The RR of 0.35 puts it in the same category as recognized genocides.

Manila (1945) shows what happens when a garrison fights to the last man in a dense city: the Japanese garrison was nearly annihilated (31.7%), but ~100,000 civilians also died. Mosul and Raqqa (~8.5) show higher RRs - the attacking coalitions used more targeted strikes while ISIS defenders took catastrophic losses. Still, ~10,000 civilians died in Mosul.

Gaza shows the widest gap between military and civilian death rates on the chart - 20.5% vs 1.28%. The civilian rate is low in absolute terms because the denominator is so large (1.5–2.2M), while the military rate is high because militant forces are small. That gap is what produces the high RRs.

What the Numbers Tell Us

The spectrum from Nagasaki (0.18) to Russia-Ukraine (339.4) spans over three orders of magnitude.

The genocide floor (RR < 1). Nagasaki, Rwanda, Nanjing, Grozny II, Hiroshima, Srebrenica. These are cases where the explicit or functional purpose was the destruction of a civilian population, or where weapons were inherently indiscriminate in civilian areas. When you see an RR below 1, something has gone categorically wrong.

The urban warfare band (RR 1–10). Grozny I, Cambodia, Manila, Dresden, the Eastern Front, Fallujah, Korean War, the Lebanese Civil War, Vietnam, Mosul, Raqqa. Urban combat inherently compresses the ratio. But the spread is meaningful: Grozny I (1.23) vs. Mosul (8.5) reflects the difference between carpet bombardment and targeted operations. Lebanon’s 15-year civil war (RR ≈ 5) sits squarely in this band - sectarian massacres and indiscriminate shelling destroyed the distinction between frontline and home.

The source-dependent middle (RR 3–50). This is where the Iraq War, the Syrian Civil War, and the Gaza conflicts live - and where the choice of source matters most. The Iraq War’s RR ranges from 3.5 (Lancet II - which would place it next to the Eastern Front) to 17.3 (IBC - solidly in the asymmetric conflict zone). Syria’s overall RR ranges from 18.8 (OHCHR’s 307K civilian estimate) to 55.3 (SOHR’s lower civilian count). For Gaza 2023–25, the range is 15.3 (if all 73k killed to date ~Jan2026 were civilians) to 42.1 (IDF estimate). Turkey’s Afrin operation (RR ≈ 29) also falls here, though the data itself is contested.

The conventional ceiling (RR > 50). The Iran-Iraq War, Gulf War, Afghanistan, Yemen, Russia-Ukraine. Militaries fighting militaries. Civilian death rates orders of magnitude lower than combatant rates - at least when you count only direct violence. Yemen’s RR of 189 belongs here by that metric, but 60% of its 377,000 deaths were from starvation and disease. The “conventional” label masks a catastrophe.

Cambodia is an outlier. Its RR of 1.76 seems too high for a genocide - but the Khmer Rouge purged their own cadres at enormous rates. Calculating RR specifically for targeted ethnic groups (Cham Muslims, Vietnamese Cambodians) drops it well below 1.

The Middle Eastern Context

The Middle East has produced some of the most varied conflict patterns on the chart. The Lebanese Civil War (RR ≈ 5) and the Syrian Civil War (RR ≈ 19–55) sit in the same region as Vietnam and Iraq, while the Iran-Iraq War (RR ≈ 92) and the Yemen Civil War (RR ≈ 189) land among conventional interstate conflicts. The same region, radically different patterns - and the reason is always the same question: who is being killed?

The Lebanese Civil War (1975–1990) killed an estimated 150,000 people over fifteen years - roughly 4.5% of Lebanon’s population. It wasn’t a war between armies. Dozens of militias, three foreign armies (Syria, Israel, the PLO), and a fractured national military fought across Beirut’s apartment blocks and mountain villages. The massacres - Sabra and Shatila, Karantina, Damour, Tel al-Zaatar - are the ones that get remembered, but the ICRC documented that most civilians died in the daily grind of shelling and sniper fire [53]. An RR of ~5 places it alongside the Korean and Vietnam Wars: conflicts where the combatant toll was staggering, but the civilian toll was worse.

The Syrian Civil War (2011–2024) is the bloodiest conflict of the 21st century by raw count. The OHCHR’s statistical estimate - 306,887 civilians killed between 2011 and 2021 alone - used multiple-systems estimation across eight data sources [51]. SOHR’s total exceeds 528,000 [52]. The Assad regime was responsible for the vast majority of civilian deaths: barrel bombs dropped on markets, chemical weapons attacks on residential neighborhoods, and an estimated 47,000 people tortured to death in detention. The overall RR of ~19 (using the OHCHR civilian count) is higher than Aleppo alone (14.4) because the nationwide data averages the worst urban sieges with lower-intensity rural fighting. Use SOHR’s lower civilian figure and the RR climbs to ~55 - a disagreement that spans nearly three color categories.

Turkey’s Afrin operation (2018) illustrates a different problem: not source disagreement, but a total data void. Operation Olive Branch - Turkey’s cross-border incursion with Syrian proxy forces - displaced hundreds of thousands of Kurdish civilians from the Afrin district. SOHR and Amnesty International documented 400–500 civilian deaths [57]. Turkey’s official position: zero. The RR of ~29 is calculated from SOHR’s figures. It sits alongside Lebanon 2006 and the Gaza conflicts, but with less confidence in the underlying numbers than almost any other entry on the chart. Amnesty also documented extrajudicial killings, ethnic displacement, and reports of chemical weapons use. The broader Turkey-PKK conflict (1984–present) has killed an estimated 30,000–40,000 over four decades, with the International Crisis Group documenting 4,310 killed between 2015 and 2018 alone - including 465 civilians in urban curfew operations that destroyed entire neighborhoods in cities like Cizre, Nusaybin, and Sur [58].

The Iran-Iraq War (1980–1988) was the longest conventional war of the twentieth century. Eight years of trench warfare, human-wave attacks, and chemical weapons - the closest the modern era has come to repeating World War I. An estimated 500,000 combatants died on both sides. The overall RR of ~92 places it firmly in the conventional range: armies killing armies [54]. But that aggregate hides the Anfal campaign - Saddam Hussein’s systematic destruction of Kurdish communities in 1988, which killed 50,000–100,000 civilians through mass executions, chemical attacks (including the Halabja massacre of 3,200–5,000), and forced displacement [55]. Calculate Anfal’s RR separately and it drops below 1 - a genocide nested inside a conventional war.

Yemen is the most misleading entry on the chart. At RR ≈ 189, it looks like precision warfare. It is not. The UNDP estimated 377,000 Yemenis dead by the end of 2021, but 60% of those deaths were from starvation, preventable disease, and the collapse of water and health infrastructure [56]. The Saudi-led coalition’s air campaign systematically struck hospitals, water treatment plants, food storage facilities, and funeral halls. ACLED documented that the coalition was responsible for 67% of all reported civilian fatalities from airstrikes. The “direct violence only” filter - which produces a clean, high RR - conceals the mechanism: you don’t need to bomb civilians directly if you destroy everything that keeps them alive. If indirect deaths were included, Yemen’s RR would collapse. The UNDP projected that if the conflict continued to 2030, 1.3 million people would be dead - 70% of them children under five.

The broader pattern here is that the Middle East spans the entire RR spectrum in a way that few other regions do. Lebanon and Syria sit in the zone of urban sectarian violence. Iran-Iraq looks like a European great-power war. Yemen looks clean until you change what you count. And Turkey’s operations exist in a data fog where the very existence of civilian casualties is contested by the perpetrator. The choice of methodology - direct vs. indirect deaths, which source to trust, how to classify combatants - shapes the result more in this region than anywhere else on the chart.

Gaza in Context

Which brings us to the conflict everyone actually wants to talk about.

Gaza 2023–25 sits at RR 15–42 depending on the source. Look at where that lands on the chart. It’s in the same band as Kosovo (13.6), the Bosnian War (13.0), and Lebanon 2006 (20.4). It is nowhere near the genocide cluster - Rwanda (0.25), Srebrenica (0.90), Grozny II (0.35). It’s not even close to indiscriminate urban warfare like Grozny I (1.23) or the Battle of Berlin (1.08). It outperforms the US-led coalitions in Mosul (8.5) and Raqqa (8.6) - campaigns that were broadly considered successful examples of urban counterinsurgency.

Now consider the operating environment. The Gaza Strip has a population density of roughly 5,750 people per square kilometer - comparable to London. Gaza City itself exceeds 13,000/km², denser than any major European city. Hamas fighters operate from tunnel networks built under residential neighborhoods, hospitals, and schools. They don’t wear uniforms. The entire doctrine is to blur the line between combatant and civilian. And in that environment, Israel is achieving combatant-to-civilian ratios that land near the top of the historical spectrum, and especially so when only considering modern urban combat, and double so against insurgents.

Then there’s the famine question. For months, the narrative has been that Israel is engineering a starvation campaign. The WHO confirmed 32 malnutrition-attributed deaths through June 2024 [59]. By end of 2024, the figure was 49. The Gaza MoH reports roughly 460 by late 2025. Those are real deaths and they matter. But context matters too. In Yemen, 60% of 377,000 deaths - over 225,000 people - were from starvation and disease caused by the Saudi-led coalition’s destruction of food and water infrastructure. The siege of Leningrad killed 800,000 civilians, mostly through hunger. Even including Gaza’s malnutrition deaths, the indirect toll is orders of magnitude below what siege warfare and engineered famine actually look like historically.

The RR data doesn’t say the Gaza campaign is clean. Thousands of civilians are dead. But the data does say this: whatever this is, it isn’t genocide, it isn’t indiscriminate, and it doesn’t pattern-match to any conflict where those words have been applied. It patterns-matches to asymmetric urban warfare in an environment specifically designed to maximize civilian exposure - and the ratio is better than most comparable operations, not worse.

Limitations

1. Population denominators are debatable. For Fallujah, I used the remaining 40,000 after evacuation - the pre-war 275,000 would give RR = 25.5 instead of 3.7.

2. Combatant classification is the key variable. For Nanjing, counting POW executions as “military deaths” jumps the RR from 0.29 to 1.96.

3. Aggregation hides variation. The Eastern Front’s 3.07 averages Leningrad (catastrophic civilian losses) and conventional steppe battles.

4. Duration bias. A 20-year war (Afghanistan) vs. a single day (Hiroshima).

5. Indirect deaths. Most figures count only direct violent deaths. Including indirect mortality would lower every RR.

6. RR doesn’t capture intent. But when every recognized genocide falls below 1 and every conventional war sits above 50, the metric is capturing something real.

Despite these limitations, the Relative Risk Ratio provides a useful framework for asking: who is bearing the cost of this conflict, and does the pattern look more like Mosul or more like Rwanda?

The Data

Raw numbers behind the charts. Click any reference number to jump to its source.

References