Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How well do cars do in crash tests they're not optimized for?

danluu.com

Post reply on HN
Print view View original

201–210 of 352 posts

Re: How well do cars do in crash tests they're not optimized for?

#201

Post by Bartweiss »

ebg13 wrote: post #95
kube-system wrote: post #90

Earlier quoted context omitted.

The big difference with VW is that they put in a mechanism that detects the test and completely changed the behavior of the car for the test.

So they _intentionally_ set themselves up to lie to regulatory agencies and consumers about real world efficiency. That honestly sounds basically the same to me. In both cases the tests are poor approximations, and in both cases someone could accidentally optimize the test, and in both cases someone did it intentionally to deceive people.

> in both cases someone could accidentally optimize the test

I think this is what I disagree with.

The water heater story is about a viable-for-market design which also optimized for the test. The equivalent for a car emissions test might be optimizing the transmission to reduce emissions at the specific speeds which will be tested. Those speeds could be sweet spots of the engine curve by accident, or they could be planned that way. I don't think that's necessarily right, but it's within the bounds of "natural" design for the product.

Instead of doing that, VW submitted something for testing which was fundamentally different from what went to market. Rather than being misleading, the test results were fundamentally irrelevant. Creating two completely different modes of behavior isn't something you could do by chance, and it means there's no real limit on how badly they could cheat.

Re: How well do cars do in crash tests they're not optimized for?

#202

Post by jhardy54 »

ghaff wrote: post #180
jhardy54 wrote: post #169

Earlier quoted context omitted.

That would make sense, except the unit is 'Fatalities per million hours'. Maybe they mean "fatalities from residential fire per million hours of living in a residence"?

I assume it means "per million hours of living." (Though I guess it could be "per million hours of living in a residence")

The problem with the unqualified "per million hours of living" is that sky diving doesn't kill more people than passenger cars.

It must be "fatalities from X during Y", and that makes sense when X and Y are the same thing (e.g. sky diving), but it seems like X and Y diverge in some of these entries.

Re: How well do cars do in crash tests they're not optimized for?

#203

Post by macleginn »

pbhjpbhj wrote: post #107
amelius wrote: post #10

Earlier quoted context omitted.

Some tips: Dress like a motor-cyclist. Wear a helmet and airbag suit. This is probably also useful when traveling by airplane. I have no data on the effectiveness, however.

People who feel safer can take greater risks, thus safety equipment can create negative returns.

This theory (known in literature as ‘risk compensation theory’) has been debunked. See some references here: https://twitter.com/whitequark/status/1264486505168482304

Re: How well do cars do in crash tests they're not optimized for?

#204

Post by amelius »

macleginn wrote: post #203
pbhjpbhj wrote: post #107

Earlier quoted context omitted.

People who feel safer can take greater risks, thus safety equipment can create negative returns.

This theory (known in literature as ‘risk compensation theory’) has been debunked. See some references here: https://twitter.com/whitequark/status/1264486505168482304

To be fair, it seems the debunking applied to people who were obliged to wear safety gear, not to people who were more cautious than average people.

Re: How well do cars do in crash tests they're not optimized for?

#205

Post by rconti »

PeterisP wrote: post #158
hangonhn wrote: post #98

Earlier quoted context omitted.

Can you explain more about the walking part? Is it because pedestrians get hit by cars a lot? Otherwise it seems walking should be incredibly safe.

Transportation safety is generally measured on a distance basis. If you compare one hour of driving vs one hour of walking, then walking is safer, but if you compare the traffic risks of driving a hundred miles and walking a hundred miles, then that's a different story.

This is also kind of silly; the average person doesn't walk 100 miles (say, a week). Extrapolating relative walking safety from 100ft of darting from your office building, across the street, to your lunch spot, out to 100 miles is not actually the same thing as (say) commuting by foot 5 miles each way for 10 days (where the per-step risk profile is probably a lot lower than the numbers we're extrapolating from).

Re: How well do cars do in crash tests they're not optimized for?

#206

Post by Sohcahtoa82 »

UI_at_80x24 wrote: post #73
mohaine wrote: post #46

Earlier quoted context omitted.

This. If driving was the only cause of death, and we did nothing but drive from birth we would expect to live to ~250 years old on average.

I enjoyed your message, and it reminded me of a completely unrelated fallacy. If 1 women can produce a baby in 9 months, then 9 women can produce a baby in 1 month.

I've always heard that one as "A Project Manager is someone who thinks 9 women working together can produce a baby in 1 month."

Re: How well do cars do in crash tests they're not optimized for?

#207

Post by orclev »

Bartweiss wrote: post #195
im3w1l wrote: post #129

Earlier quoted context omitted.

Many important differences. 1. Actively detecting test and behaving differently. It's like stealing a test vs teaching to the test. 2. Lower stakes. Health issues are much more serious than inefficiency. 3. It affects the buyer. It's more acceptable for the buyer to be cheated than everyone around them. 4. People could have created these layers by accident. Favouring those who got lucky is unfair. Honestly I think ba…

This is omnipresent even where regulators aren't involved: every graphics card benchmark out there is 'manipulated' relative to real world performance. At this point it's so universal that I don't think anyone is even fighting it - as long as everyone games benchmarks roughly the same amount, the relative scores stay usable. Your point about fairness and passive design is the one that makes me view these cases differ…

It's interesting you mention graphics cards because that very behavior has lead to the gaming community favoring benchmarks derived from a handful of current gen games min/max/avg FPS over so called synthetic benchmarks. It only took a handful of instances of companies baking in "benchmark" modes that get triggered when certain benchmarks are detected for people to start discounting those benchmarks in favor of more organic measurements.

Re: How well do cars do in crash tests they're not optimized for?

#208

Post by dredmorbius »

bananamerica wrote: post #197
chickenpotpie wrote: post #102

Earlier quoted context omitted.

Also https://en.wikipedia.org/wiki/Goodhart%27s_law

How did I know this linked to some kind of general principle with a cool short name even before reading or opening the link?

Well, there are several such:

https://en.wikipedia.org/wiki/List_of_eponymous_laws

Though I'm not aware of a law stating that for any given principle there is an existing eponymous law.

Re: How well do cars do in crash tests they're not optimized for?

#209

Post by ghaff »

jhardy54 wrote: post #202
ghaff wrote: post #180

Earlier quoted context omitted.

I assume it means "per million hours of living." (Though I guess it could be "per million hours of living in a residence")

The problem with the unqualified "per million hours of living" is that sky diving doesn't kill more people than passenger cars. It must be "fatalities from X during Y", and that makes sense when X and Y are the same thing (e.g. sky diving), but it seems like X and Y diverge in some of these entries.

Though even skydiving has a very different time/risk profile than, say, mountain climbing (however you care to define). If skydiving time is measured only by time from leaving the plane to hitting the ground, you're measuring a fatality rate against a pretty short interval of time. (Same with scuba diving.)

Whereas other activities that may have similar risk profiles for typical "active participants" may have much different rates relative to the time spent actively participating in the activity.

There's no single metric that really represents a correct view.

Re: How well do cars do in crash tests they're not optimized for?

#210

Post by duskwuff »

baxtr wrote: post #76
shoes_for_thee wrote: post #43

Earlier quoted context omitted.

This might have some useful info for you... https://www.cyclehelmets.org/1026.html death rates per million hours of activity

Thanks! That’s really a great source. I am so glad I asked. Highlights: - Fishing seems to be quite dangerous!!?? - Swimming is less dangerous than living (all other causes)!

"Fishing" might include "working on an industrial-scale fishing boat", which is a notoriously dangerous line of work.
Post reply on HN