How rare is a rare death?

The Bizarre & Rare section is the most enjoyable part of this site and, by a wide margin, the least reliable. That is not an accident of sloppiness. It is a structural problem with the whole category, and it is worth explaining before going through the individual claims.

The column has twelve counters. The annual figures below are the ones printed on those cards now. Where a card and the evidence disagree, the disagreement is spelled out.

Why rare-death statistics are so bad

Common causes of death are counted through civil registration1: a certificate, an ICD code2, a national total. The system is imperfect, but it is a system.

Rare causes fall through it. There is no ICD code for "struck by a falling coconut". A death like that is coded as blunt force trauma3, or simply as a falling object, and the coconut vanishes. Anyone wanting a global figure cannot look it up. They have to gather news reports, stretch one local study across the world, or repeat a number someone else produced that way.

News reports have a specific and severe bias. They count what gets written up, and that depends on novelty, language and how dense the local press is. A shark attack in Australia is reported worldwide. An identical death in a country with less international coverage may never appear in any dataset. These figures therefore undercount unevenly, and once published they are repeated indefinitely, with nobody going back to the original count.

Tier one: reasonably documented

Fatal shark bites (~10 a year). Genuinely well tracked. The International Shark Attack File at the Florida Museum of Natural History has catalogued unprovoked4 bites for decades. Fatalities in that file are usually in the single digits. A recent five-year average was about six unprovoked deaths a year. Across 2010 to 2019 it logged 54 fatal attacks, about five or six a year. The year 2023 reached ten unprovoked deaths, and 2024 fell to four. The Ledger's 10 sits at the top of that recent range, not in the middle of it. The card says bites, because attacks are the wider word, and the second line is the 2024 count of four unprovoked deaths, not a decorated last year. Sharks are unusually well counted precisely because they are frightening enough for someone to have bothered.

Lightning strikes (~4,000 a year, and probably more). Lightning deaths are well documented where weather services are good, and poorly documented where most of them happen: agricultural regions, outdoors, with little shelter. Adding up the countries that publish a multi-year count produces just over 4,000 deaths a year. The Ledger rounds that to 4,000. The sum leaves many countries out. Estimates of the true global toll, made by extrapolation5 from the rates that are known, run from about 6,000 to about 24,000. The Ledger's 4,000 is the documented floor, not a cautious guess at the real number.

Lift failures and roller coaster failures have left the dashboard. Both were United States figures wearing a global-sounding noun. The lift count was about 30 deaths a year in the 1990s and 2000s, lifts and escalators together, and about half were people working on or near the lift. The roller coaster count was 40 associated deaths from May 1994 to May 2004. Of the 29 riders, 18 died of a medical event, including intracranial haemorrhage6, and 11 died of a fall or a collision. The cards said "failures". The counts were not that. Their places are taken by series that name the country, or the register, on the card.

A register, with the country on the card

Commercial aircraft, people onboard (~200 a year). This replaces the hot-air balloon card, whose 8 a year was a round guess. IATA’s 2025 safety report counts onboard fatalities on commercial passenger and cargo aircraft with a maximum take-off weight of at least 5,700 kilograms. The five-year average for 2021 to 2025 is 198. The card ticks 200. The years do not sit on that mean: 72 onboard deaths in 2023, 244 in 2024, 394 in 2025. The second line on the card is that 2025 count. It is a real year, and it is allowed to disagree with the mean. General aviation is a different, larger toll, and it is not this card. The longer essay is People on board.

Bee, wasp and hornet stings (~70 a year, United States). The US Centers for Disease Control and Prevention counted 788 deaths from 2011 to 2021 with underlying cause X23, hornet, wasp and bee sting. That is an average of 72 a year. The card rounds to 70 and says United States. It is not a world total. The essay is Coded X23.

Civilian skydiving deaths (~15 a year, United States). The United States Parachute Association counted 9 civilian fatalities in 2024, from about 3.88 million jumps, and 16 in 2025, from about 3.47 million. The card ticks 15, and the second line is the 2025 count. It is the United States, and it is deaths, which is the correction to the roller coaster card it replaced. The essay is Nine, then sixteen.

Pufferfish poisoning, Japan (~1 a year). Japan’s Ministry of Health, Labour and Welfare publishes a food-poisoning series for fugu. Across the recent decade the deaths are zero or one a year. The 2025 table shows 10 incidents, 15 patients and no deaths. The card ticks 1, which is the size of a bad year in that series, and the second line is 2025’s zero. It is Japan only, and it is a sliver of the poison counter, which is accidental swallowing or inhalation worldwide. The essay is The liver, at home.

Tier two: weakly sourced but plausible

Scorpion sting deaths (~3,000 a year). Reviews of national health data put the world toll at roughly 2,600 to 3,250 deaths a year, against well over a million stings, and they say the deaths are under-reported. The card rounds to 3,000 and says the estimate is short. A sting is not the poison card. In Mexico, widespread antivenom took a national toll that had been about 2,000 deaths a year down to under 50. The essay is Under fifty.

Snow avalanche deaths (~250 a year). The European Alps have lost about 100 people a year to avalanches across recent decades, and that series is the solid part. The United States has averaged about 27 deaths a winter over the last ten winters. A world figure of about 250 is the estimate repeated in the avalanche literature, and high mountain Asia is thinly counted, so 250 may be low. The card says so. It replaces falling icicles, which were a round 100 with a Moscow anecdote and no register. The essay is A hundred in the Alps.

Hippopotamus deaths (~500 a year). Hippos are genuinely dangerous: territorial, fast in the water, and often met by people fishing or travelling by boat. The figure of about 500 deaths a year is the one repeated in "deadliest animals" lists, and it does not come with a census. Our World in Data, reviewing the national records that do exist, notes 20 deaths in Zimbabwe from 2016 to 2022 (about three a year) and about 11 a year in Zambia, which holds about a third of the world's hippos. From figures of that kind it uses a global total nearer 50 than 500. Treat the Ledger's 500 as a high guess that has been copied.

Fatal selfie attempts (~380 a year). Selfie deaths have been counted from news reports, which miss cases and are not a register. A 2018 review in the Journal of Family Medicine and Primary Care found 259 deaths worldwide from October 2011 to November 2017. That is just over six years, about 42 a year across the span. The years were not steady: the same paper records 98 deaths in 2016 and 93 in 2017. A later news-report count, from 2008 to mid-2021, found 379 deaths in thirteen and a half years, with 68 in the busiest year, 2019. The Ledger's 380 a year is not a rate either study reports. It is the size of that later cumulative total, read as if it happened every year.

Vending machines, falling icicles and hot-air balloons have left the column. The vending figure was a United States count from 1978 to 1995, about two deaths a year, and probably high for today. Icicles were a round 100 with no register. Balloon crashes at 8 a year were a guess; the aviation card above is the register that replaced them.

Tier three: folklore

Falling coconuts (~150 a year). The famous claim that coconuts kill 150 people a year, usually introduced as "more than sharks", has no basis. It grew out of a 1984 paper in the Journal of Trauma by Peter Barss, on injuries treated at one hospital in Papua New Guinea over four years. In that series, 2.5 per cent of trauma admissions were people struck by falling coconuts. The paper describes a handful of local head injuries, including two people who died in their village. It does not estimate a world total. The "150 a year" figure came later. In 2002 the director of the International Shark Attack File, at a press briefing, put falling coconuts at 150 deaths a year and called that fifteen times the shark toll. Fifteen times ten is 150. It is a shark comparison with the coconut number filled in afterwards, and it has been repeated ever since. There is still no global count.

Right-hand gear accidents have left the dashboard. The claim that 2,500 left-handed people die each year using right-handed equipment was not a published count. It descends from papers in 1988 and 1991, by Diane Halpern and Stanley Coren, which argued that left-handers die about nine years younger than right-handers. Those papers suggested accidents in a right-handed world as one possible reason. They did not count 2,500 such deaths, or any annual total of that kind.

The lifespan finding itself did not survive scrutiny. The leading explanation is a cohort artefact7. For much of the twentieth century, left-handed children were made to write with their right hands, so older generations report as right-handed at much higher rates than younger ones. Compared that way, left-handers look short-lived, not because they die early, but because fewer old people were recorded as left-handed. Later studies have not found a meaningful gap in lifespan.

Spontaneous human combustion (~1 a year). Not a recognised cause of death. Forensic8 investigators generally explain the handful of celebrated cases through the wick effect9: clothing acts as a candle wick and body fat as the wax, so a small external flame, such as a cigarette or a heater, can burn for hours. The body is badly destroyed while very little of the room is scorched, which is the detail that made the phenomenon seem supernatural. There is no evidence of a person igniting with no external source. One a year is a storytelling rate, not a count.

Why we get this so wrong

These figures circulate because risk intuition runs on availability10. We judge how likely something is by how easily an example comes to mind, and vivid, novel, heavily reported dangers come to mind very easily.

The Ledger's shark card says about 10 deaths a year. Its cardiovascular11 counter uses 18.6 million. Divide the second by the first and the ratio is 1.86 million to one. Using the shark file's longer average of about six a year instead, 18.6 million divided by six, the ratio is 3.1 million to one. Almost nobody's sense of danger reflects either figure. The risks that actually dominate a life arrive with no story attached: a poor diet, a decade spent sitting, untreated high blood pressure, a cigarette habit.

Which is, in the end, the joke the Bizarre & Rare section is making. The coconut card says 150 deaths a year, and its since-arrival counter is accelerated 160 times, so that something moves during a visit. Even then it is slow. At the cardiovascular counter's true rate, 18.6 million divided by 150 and then by 160, it moves 775 times for each tick of that coconut counter. You are watching the coconut, and you almost certainly did not look at the other one.

More from the Ledger

Last reviewed: 6 October 2026