Journal: News & Comment

Damn lies, statistics, and fatality rates

Permalinks to this entry: individual page or in monthly context. For more material from my journal, visit my home page or the archive.

My post yesterday about the hazards of being an astronaut brought a thoughtful response from my former UBC colleague, the excellent photographer Alastair Bird:

Is there a minimum required sample size [for dangerous jobs] and are you speaking overall? It would appear that being an astronaut currently is the most dangerous job. In WW�II, I think half of all the sailors on German U-Boats were lost and about a third of all Allied bomber crews (RAF and RCAF) were lost. Those were alarming statistics, but they were actively being shot at -- I think there were few accidental deaths. And I think that until recently (when large groups of climbers starting climbing it) I thought about 1 in 20 people who attempted the summit of Everest died.

The Everest numbers he mentioned (5%) roughly match the 4.3% high-mountain stats I quoted, but I had no idea the fatality rate was so high among some World War�II military groups. Indeed, Alastair was underestimating it: U-boat crews had a 63% fatality rate, or 63,000 per 100,000, by the criteria I used yesterday. That's about 500 times more dangerous than logging, and close to ten times as dangerous as being an astronaut. But again, people were actively trying to kill them. It's pretty hard to compare that kind of work -- fighting military, gang and organized crime members, and so on -- to jobs in which someone's not supposed to die as a routine part of achieving your goals. More realistic than U-boat crews would be submarine crews as a whole, I suppose.

However, I suspect my astronaut (and mountaineer) fatality statistic comparisons may fail for a more critical reason: the mountaineering and astronaut numbers are historical averages, while the logging etc. numbers are one-year snapshots (I think). So they may not really be directly comparable.

One-year snapshots for the space program (and extreme-altitude mountain climbing) are difficult, though, because so few people are involved and the death rate varies so widely -- long periods of zero punctuated by spikes of horribly high fatality rates. And getting historical average rates for more common industries seems hard to do, since most of the statistics seem to be annual. However, because of the larger numbers of people in those industries, the historical averages and snapshot rates should be similar enough, especially for the orders-of-magnitude differences we're talking about.

Alastair continued:

I can't help but think (this is perception, not necessarily statistical reality) that for the guys who are hard-core [high-mountain] climbers, death somehow always seems to catch up with them. Anatoli Boukreev [...] who was a big part of [the 1996 Everest expedition where many died] was killed on Annapurna a few years later. I think several of the guys who managed to get off of Everest and not die have been killed in subsequent climbing accidents, and even our own Jim Haberl, from Whistler was killed in a pretty innocuous avalanche in Alaska a couple of years ago.

Perhaps the really heavy-duty, repeat climbers do have a higher death rate. But how many of them are there, and how do you distinguish them from the less hardcore climbers, or those who climb all the time at lower altitudes? We can always slice the numbers (or lump them together) until they become meaningless. Those who helped contain the Chernobyl reactor as it melted down in 1986 have a high death rate, for instance, but is that more meaningful than the numbers for nuclear plant technicians worldwide, which are probably pretty low? Is the worldwide average death rate from AIDS more or less useful for identifying trends and treatments than separate rates for industrialized North America and sub-Saharan Africa?

How about numbers for mountaineers in general, not just those who go to the extreme altitudes of the Himalayas? What if we lump astronauts in with everyone who's ever flown in a heavier-than-air craft above the ground? Then things look pretty safe. My statistics lie no more than anyone else's -- they just lie differently. At least with the numbers I roughed out, whether someone is an astronaut (or cosmonaut) and has been into space at all -- once or repeatedly -- is relatively clear.

There is another contrast: mountaineers often die because of spur-of-the-moment, individual decisions, sometimes entirely alone. They step in the wrong place and fall, or choose to make for a peak too late in the day, or misjudge weather conditions. Astronauts work in the cushion of years or decades of planning and training by thousands of the smartest people in the world, with budgets in the tens of billions of dollars. But they still die more often. That shows how dangerous their jobs are, and why private spaceflight is likely a long, long way off.

I remember a fellow on the CBC speaking about risk and disease [and] the interviewer asked him how anyone could actually live to their expected life expectancy with all this disease and death. His response was, "Wear your seat belt. Quit smoking. Get more sleep. Don't drink too much, but have a drink or two regularly. Quit your stressful job and take regular vacations." He said that doing this would take care of the vast majority of the expected mortality -- the rest was just up to chance and that everyone would have to accept the reality that awful things happen to regular people.

The vast majority of people do live into old age, and then die of heart disease, cancer, pneumonia, or other "natural causes." And driving remains pretty dangerous, as "regular people" things go. The fatality rate is around 22 per 100,000 drivers, while the rate for scuba diving (as an example that may seem dangerous) is about 2 in 100,000 (see June 29, 2002 on this scuba page).

Everyone knows someone who's been in a car crash (if not ourselves), and many of us know someone killed in one. But we still drive. If every car trip were as carefully planned and rigorously executed as each space shot, no one would ever die in a car crash. And in real life, some people have no collisions for years, then have two (even not their fault) within months of one another. That demonstrates how randomness appears, sometimes in clusters that seem related, but aren't. Like having two groups of seven people die last Saturday in completely unrelated events.

 |