Hacker News Reader

Comment by wesselbindt | Hacker News Reader

I'm not sure why looking at the incidents per trip would be preferable over looking at the incidents per mile. If I think about comparing two activities when it comes to safety, I feel like I'd want to look at the number of incidents per second of performing either activity.

EDIT: here's a thought experiment. You're taking a trip to the grocery store. You have two routes in mind, the direct one, which takes 5 minutes, and the detour via the North Pole, which will take weeks. One of these trips is safer than the other.

View on HN

> I'm not sure why looking at the incidents per trip would be preferable over looking at the incidents per mile.

Personally when enter a vehicle and start a trip, I'm interested in arriving at the destination alive, regardless of distance. If something bad happens, whether I die 1km into it, or 10.000km into, is quite pointless. So incidents per trip probably conform better to how people evaluate travel risks.

That might be true for most and is a reasonable way of looking at it.

But if someone is going on a 10,000km trip and trying to choose between modes of travel, the mode with the fewest incidents per mile will be the safest way.

Exactly, so when I get in that car, I want to know the probability of dying. How do you compute this? By taking the number of crashes per kilometer, and multiplying by the number of kilometers you'll drive for. A trip of 5000km is less safe than a 5km one. That's just common sense. Similarly, when comparing flying versus driving from Amsterdam to Paris, you are, as you say, interested in which method is most likely to get you to your destination alive. So you take the number of incidents per kilometer for planes and cars, and multiply them by the distance between Amsterdam and Paris. The interesting figure really is incidents / kilometer, that's just basic math.

However the risks aren't necessarily proportional to miles travelled.

So for example with flights - I'd imagine the most dangerous parts are the take-off and landing - which happen once per trip no matter how long you fly for inbetween.

So for flights incidents / kilometer are perhaps misleading.

Fatalities per trip is similarly misleading because of the number of extremely short trips people make in cars.

If you really start including risk factors you'll never stop, the risk is proportional to many other factors as well:

1) Time of day: Time of day factors in probably for both flying and driving at varying degrees

2) Where: In the US, which state you're driving in matters significantly for driving fatality rates

3) Day of week: Driving on weekends is more dangerous than driving during the week, doesn't seem to affect flights

4) ATC is a variable for flight accidents but not drivers

5) Flights might actually carry more risk per mile in certain situations: Flying over the oceans, eg

A truly correct apples-to-apples comparison would be so hyper-specific as to be useless.

The other key point when comparing car stats to flights is car stats include pedestrian/cyclist and motorcyclist deaths which make up perhaps around half of the deaths.

Yet I suspect the people in twin towers weren't included in airline death stats.

Sure there isn't a perfect single metric - just pointing out that the one the airlines prefer appears to be biased towards air-travel.

The other factor that is often ignored is the issue of control. As a passenger of an airline you have very little impact on the chance of an accident - you are just an accident statistic.

Whereas if you are driving, your own driving behaviour, choice of car, route etc puts you more in control of the chances ( though of course by no means completely ).

This has a big impact on the perception of risk ( as more than average think they are better than average :-) ).

> taking the number of crashes per kilometer, and multiplying by the number of kilometers you'll drive for. A trip of 5000km is less safe than a 5km one.

When comparing a difference of 1000x, sure its probably safer.

Your premise is wrong though, it isn't common sense to take a generalized statistic like deaths per km and extrapolate that to be the correct estimate for any particular trip.

The statistic loses all context of differences in vehicles, road conditions, time of day, specific roads, etc. The stat is a great example of how a statistic seems meaningful to get a big picture but is completely useless to us as we can't make any decisions based on a number that is so generalized as to lose all context and meaning.

Common sense means taking the less risky option.

Assuming that the survival rate is constant as a function of trip length, the empirical estimate of the risk is fully determined by the number of deaths per distance.

In this case, they’re not fully determined by distance.

Airline travel is much riskier in takeoff and landing phases. A bunch of 200 mile flights is riskier than a 3500 mile trans-con trip.

Car travel is much safer on controlled access highways than 40 mph surface streets.

> Assuming that the survival rate is constant as a function of trip length

That's an assumption that doesn't hold up, at least that's the argument I was trying to make above.

Generalizing so much data down to a single statistic makes the number useless. Almost all context is lost, really the only context kept is mode of transportation and distance traveled. That isn't enough to help anyone make a decision for any particular trip.

So if I take two 100km trips by car, one takes 30 minutes and the other 2 hours both are equally safe?

Ok, so go the to nearer amenity and try to live closer to work?

A frequent flier is still more likely to die in a car accident than airplane crash. Hell, even airline pilots are more likely to die driving to work than in their planes. You have to get to risky GA practices before the risk of dying in the plane dominates the normalised terror that our roads are.

How often do you fly and how often do you drive by car?

> How often do you fly and how often do you drive by car

I think I fly on average a bit more than once every two weeks. I'm in a car about every day, mostly for short trips--in the last month I probably spent about as much time in a car as in a plane. Pilots probably fly more hours than they drive.

You really have to fuck with the statistics to get driving to come anywhere close to commercial aviation in terms of safety, they're so far apart.

If you take the area where you drive into consideration the death numbers change, the same is valid for the drivers behaviour.

You can't influnce that when you fly.

> If you take the area where you drive into consideration the death numbers change, the same is valid for the drivers behaviour

Source?

> You can't influnce that when you fly

Whether you can influence a variable doesn't change its order of magnitude. Unless you're driving on a closed track at low enough speeds that if you pass out you'll be fine, you're safer flying than driving.

> Source?

Look at accident reasons in the statistics. Velocity is a big part.

Changing the variable doesn’t change the overall order of a magnitude but the personal one.

If you don’t DUI your risk of dying because of DUI is lower

Unsafe air travel has been regulated out of existence in the best regions. Piloting a small airplane in a random country is still dangerous though. Unfortunately it is hard to avoid doing the equivalent in a car.

In theory it would be possible to create a safe driving car that would avoid unsafe roads, conditions, mental state, amount of driving, driving practices, speeds, distances etc. But a lot of the time it would just end up not being compatible with the road system and owning a car. And it still wouldn't totally prevent an accident, especially not caused by anyone else.

> Piloting a small airplane in a random country is still dangerous though. Unfortunately it is hard to avoid doing the equivalent in a car.

As someone who does both, the former is still controlled in a way we just don't do for cars.

Note that the moment we go to countries that are actually willing to take people's driver's licenses (and put them in jail when they drive without one, the way you'd expect to be treated flying a plane unlicensed), the safety statistics change. The only way we're getting that in America, however, is with self-driving cars.

Personally if I'm about to go on a 500 mile trip and choosing between modes of transport on safety grounds, I'd pick the one with the highest chance of surviving that trip, which ceteris paribus is the one with the fewest incidents per mile, and not make the error of assuming that the risk of a 500 mile car journey is equal to the risk of the average 10 mile car journey the per-trip stats are based on.

(Complications like takeoff and landing might actually make a 10 mile commercial flight riskier, but that isn't an actual option. The 500 mile flight is going to work out safer in almost all circumstances though)

Travel can be necessary or discretionary. The per-mile safety is the relevant number when the travel is necessary, but if it’s discretionary, then I might choose to just stay home if the numbers (price, time commitment, missed events back home, death risk…) don’t seem that great.

Sure, but if you're evaluating the risk of a discretionary trip the crude per mile safety of the transport mode (multiplied by the specific trip mileage and adjusted by any other risk factors you're aware of) is still the correct starting point for estimation and the crude per trip estimate across all journeys undertaken by that mode of transport raised upthread still unambiguously the wrong one (especially if there are vast differences in average trip lengths like between aeroplanes and cars)

In most instances, people tend to budget travel by time rather than distance.

All things equal, people tend to budget about (one-way) 10 minutes for shopping, 20 minutes for commuting, and about 3--12 hours for holiday travel.

By foot, that translates to about 1/2, 1, and 6--36 miles (at 3 mph).

By bike, about 2.5, 5, and 45--180 miles (at 15 mph).

By auto, about 5, 10, and 90--360 miles (at 30 mph, typical aggregate speed).

By high-speed rail, 25, 50, and 450--1,800 mi (averaging 150 mph).

By plane, the shorter trips don't make much sense, but 1 hour travel will get you 150 -- 500 miles, 6 about 3,600 mi (continental distance), and 12 hours > 7,000 mi (intercontinental).

Numbers are rough, but for ballpark comparisons should work.

People didn't begin travelling continental / transcontinental distances for holiday or business until air travel was widespread, cheap, and safe. Where exceptional travel was absolutely required it was undertaken, though until the age of steamships, very very few people undertook ocean voyages of more than a few days' duration. Crossing oceans took weeks, if not longer.

It’s more than a planning topic. A plane trip is atomic, you can’t cit it in pieces (it just becomes “more trips”). Once on a plane you are there till the end of the trip and flying 1mi or 10000mi is still one trip. The length of the trip has very little statistical impact on the outcome. Statistically most crashes are on takeoff or landing, length of flight doesn’t impact this.

The point is that whether you crash in the first mile or the last mile, you still crashed on a trip. Since you can’t count 9999mi safe, and 1mi deadly then you’re realistically doing statistics on trips, then multiplying by miles but this just inflates the numbers and muddies the waters.

The multiplication by miles isn’t obviously wrong but it’s definitely misleading. It’s the kind of thing that sounds relevant but isn’t really.

But people generally don't take airplanes for groceries or commuting.

If you are going on holiday, you have a time budget for reaching that particular destination. With a car you would maybe choose places within a drive of 3 to 11 hours (at highway speeds), whereas a plane would maybe be between 1 hour and 16 hours. But then a longer road trip might also be more dangerous.

There are people who commute by air, usually the very rich and for shorter distances by helicopter. Speeds are lower than fixed-wing aircraft, but landings can occur in more places. Trips start in the 10s of miles range, which at airspeed is on the order of the 5--10 minutes I'd put at my lower-range of time budgets.

Current commercial helicopter transfer service from Manhattan to JFK (about 15 miles) is scheduled for about 5 minutes:

<https://www.blade.com/p/jfk>

In your second example, you're missing the point. People tend not to take longer road trips out of a safety budget but a time one. With a limited vacation time budget, time spent travelling (unless that's the point of the vacation) is lost. Auto-based vacations tend to be closer to home.

It's the direct issue in decision making. You ask "is this trip safe for me to take?", and have example of "it's the same risk as a single car trip to work"

It depends. Are you flying because you need to get to the destination? You will have to do the miles so use per mile. Are you flying for vacation? Is the alternative to just drive to a nearest lake/mountain and vacation there? Then use per trip

They both have their place when used in the appropriate context.

With aircraft, takeoff and landing are the riskiest activities, for example. So per trip reporting is relevant. The scope of what an incident is matters as well.

Reply on news.ycombinator.com