Beyond Crashes: Standard Flight Safety Estimates Underestimate Deaths by Up to 160x by Ignoring Clots, Radiation, and Pathogens

All things considered, flying is only 0.4 to 5 times safer than driving. Plus: two low-hanging fruit public policies to reduce the risks of flying and driving.

On the one hand, direct immediate deaths due to crashes are much lower per mile traveled by air than by road: 0.04 deaths per 10^8 passenger-miles for commercial airlines vs. 1.3 deaths per 10^8 passenger-miles for cars. This is well known but incomplete. Air travel exposes you to other dangers not included in that statistic: clots, radiation, and pathogens. Is air travel still safer when we quantify all those risks? Let’s do one of my favorite things: a back-of-the-envelope calculation. I don’t claim this will be precise, but putting some good-faith numbers on it is better than having no numbers.

Thromboembolism: A type of vessel blockage often caused by sitting still for too long. The incidence is approximately 20 per million passengers on a long-haul flight, of which about 1/6 are fatal, resulting in a death rate of 3 per million passengers on a long-haul flight to Australia. I’m estimating 4000 miles as the average distance traveled, which is the distance from Sydney to the south end of China. This adds 3 deaths per 4 billion passenger-miles, which equals 0.075 deaths per 100 million passenger-miles. However, the risk is very nonlinear—shorter flights will have far fewer and longer flights will have far more embolisms per passenger-mile. It is essential to avoid sitting continuously for long periods. In a car, you’re spending six times more time to travel the same distance on the highway, but in smaller increments because you can stop to rest whenever you like (and you are forced to get gas at least every 8 hours), so I’ll estimate it’s three times higher, or 0.225 deaths per 100 million passenger-miles due to thromboembolism in cars.

Radiation: The radiation exposure during a flight is approximately 0.004 mSv per hour. The cancer risk is 5% per sievert, with about half of those cases being fatal within 10 years. Therefore, one sievert corresponds to 0.025 fatalities. One hour of flight exposure equals 1/250,000 sievert, resulting in 10^-7 fatalities per hour of flight. At a typical cruising speed of 500 mph, this translates to 0.02 fatalities per 100 million passenger-miles.

Driving a car slightly increases the risk of skin cancer through UV exposure, but quantifying this is complex. Let’s assume a linear no-threshold model, not because it’s necessarily correct, but because it simplifies the math. If five bad sunburns double your risk of melanoma and one bad sunburn equals 12 hours of direct sunlight, then 60 hours of direct sunlight without sunscreen doubles your risk of melanoma. Car glass blocks half the UV, the sun only hits the side window half the time, and clothing covers 80% of your skin in the car. Therefore, 1200 hours in a car doubles your risk of melanoma, increasing the absolute lifetime risk of death from melanoma from about 1 in 1000 to about 2 in 1000. At 80 mph, this equates to 1200 x 1000 x 80 = 96,000,000 miles per extra death from melanoma. Let’s round it to 1 death per 100 million passenger-miles.

If we multiply that by the approximately 5 trillion highway passenger-miles per year in the US, we get 50,000 deaths, which is five times the total deaths from melanoma, so this can’t be correct. If we assume only half of total melanomas deaths are due to UV exposure while driving, the rate is 0.1 death per 100 million passenger-miles. In a plane, the windows are made of materials highly effective at blocking UVA/UVB, unlike car side windows.

Pathogens: I can’t find any good studies on this, but my estimate from personal experience is that the risk of getting sick from a medium-distance flight (2000 miles) is between 1 in 20 and 1 in 4. Even pathogens that don’t immediately kill you can weaken you in ways that increase the risk of dying from other causes. They can also increase your risk of cancer. Direct, immediate deaths from seasonal flu are 1 per 1000 infections, so my lower bound on the IFR (infection fatality rate) of a generic virus caught by a generic air passenger, including eventual and indirect effects, is 1/10 of that, i.e., 1 per 10,000 infections, but it could be as high as 1 in 2000. The lower bound of 1 death per 10,000 infections times 1 infection per 40,000 passenger-miles equals 1 death per 4 * 10^8 miles, or 0.25 deaths per 100 million passenger-miles. Upper bounds have 5x the infection risk and 5x the IFR for a combined 6.25 deaths per 100 million passenger-miles. My estimate from personal experience is that the risk of getting sick from driving the same distance is at most 1/10 as much because your only potential exposure is brief stops at gas stations, while planes are full of coughing passengers packed together like sardines. Let’s briefly sanity check that upper bound. There are about 100,000 deaths from infectious diseases per non-pandemic year in the US, but these are only the direct and immediate deaths, not counting those that increase your risk of cancer or weaken you, making it easier for something else to kill you later. 6.25 per 100 million passenger-miles times 7 billion passenger-miles equals 6.25 * 7000 = 43,750 deaths from air travel infectious disease, including indirect and delayed ones. If including indirect/delayed mortality doubles mortality, and 1/4 of all infectious disease transmission to adults occurs in planes/airports (my experience seems to support this), then 6.25 per 100 million passenger-miles would be about right.

There are some known unknowns that favor flying. Non-fatal car crashes also cause significant morbidity that reduces lifespan, although I don’t have a way to quantify this. Additionally, getting some exercise by walking through the airport is beneficial.

Air pollution exposure is worse when you’re driving, and I can quantify that. In one study, truck drivers had an approximately 0.3% increased absolute risk of lung cancer, and lung cancer has a 10-year fatality rate of 90%. If we model “being a truck driver” as spending an extra 10,000 hours on the road, that’s 0.27 deaths per million miles on the road, or 0.2 deaths per 100 million miles driven. Sanity check: the total lung cancer deaths in the US are 127,000 per year, divided by 5 trillion passenger miles, equals 1.8 per 100 million passenger-miles if all lung cancer was attributable to driving. It seems very plausible that 1/9 of all lung cancer is attributable to driving, so I endorse 0.2 deaths per 100 million passenger-miles.

Putting it all together:

Planes: 0.04 (crashes) + 0.075 (clots) + 0.02 (radiation) + 0.25 to 6.25 (infections) = 0.385 to 6.385 deaths per 100 million passenger-miles

Cars: 1.3 (crashes) + 0.225 (clots) + 0.1 (sunlight) + 0.025 to 0.625 (infections) + 0.2 (air pollution) = 1.85 to 2.45 deaths per 100 million passenger-miles.

Given these uncertainties, the ratio of fatalities per mile could be anywhere between 5x and 0.4x. So I’m only around 66% confident that flying is safer, if we assume a log-uniform distribution over that range.

I admit I am biased about the pathogens because I have a very high disutility for getting sick. If my disutility is $1000 per day of illness, and the average sickness duration is 5 days, then the risk of getting sick from a 2000-mile flight, which could be as high as 1 in 5, makes that risk much more expensive in expected value ($1000) than the actual airfare ($200). There’s a high density of passengers combined with a ticketing incentive structure that strongly discourages anyone from delaying a flight on account of being sick, so every flight has many passengers coughing all over each other, and it’s gross.

People who are actively coughing should be required to wear masks in airports and on planes. This could probably prevent around half of the 0.25-6.25 deaths per 100 million air passenger-miles. With 700 billion air passenger-miles per year in the USA, this equals 875-21,875 lives saved per year at a cost of approximately $5 per flight, times 10 million flights per year, totaling $100 million, or about $50,000-$2500 per life. This is quite a low-hanging fruit, perhaps comparable to malaria bed nets in purely dollar terms. However, people don’t like masking. How many person-mask-hours does it take to save a life in this situation? An average flight duration of 2 hours times 10 people masking per flight times 10 million flights equals 200 million person-mask-hours to save 875-21,875 lives. That’s 10,000 hours of masking at the low end to 200,000 hours at the high end to save a life. If the disutility of masking is $5 per hour, that’s $50,000-$1 million worth of inconvenience to save a life, which is still worth it in my opinion. A similarly low-hanging fruit is requiring full UVA/UVB protection in car windows to save up to 5,000 lives per year at a cost of about $100 per car sold, or $272,000 per life. Western liberal democracies generally use $1 million-$10 million as the value of saving a human life in their cost-benefit analyses, so masks and UV protection are low-hanging fruit that can save lives far more efficiently than the marginal dollar these governments spend.