How well do cars do in crash tests they're not optimized for? • Hacker News Board

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

#191

Post by throwaway0a5e » Tue, Jun 30, 2020, 7:08 PM UTC

nostrademons wrote: post #29

nradov wrote: post #11

Earlier quoted context omitted.

The most relevant research on car brands and models is probably the IIHS driver death rate data. It's not perfect since the populations are self selecting (cars aren't assigned to drivers at random) but within a particular vehicle class if you see a huge difference in death rates then that's probably meaningful. https://www.iihs.org/ratings/driver-death-rates-by-make-and-...

An interesting instance of the self-selection problem seems to show up if you select small 4-door 2017 cars. The Nissan Juke 2WD has a death rate of 12; the 4WD version of the same car has a death rate of 65. I have trouble believing that there's a significant difference in the cars themselves, so it seems likely that people who would buy a 2WD small SUV are a significantly different demographic than those who would…

>so it seems likely that people who would buy a 2WD small SUV are a significantly different demographic than those who would buy a 4WD small SUV.

Exactly. Nissan and FCA will finance anyone with a pulse and they make cars at low price points. The Juke is just about the cheapest AWD crossover so you get a statistically unflattering clientele driving it. You can see the reverse situation with the Dodge Journey. It is known for being the cheapest people hauler on a per-seat basis and also has a very unflattering set of customers. The 2wd is statistically more dangerous than the 4wd presumably because the riskiest buyers don't care how many driven wheels it has they just need something with a lot of seats for cheap to haul around their massive family.

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

#192

Post by DavidPeiffer » Tue, Jun 30, 2020, 7:09 PM UTC

GloriousKoji wrote: post #176

DavidPeiffer wrote: post #31

Does anyone know of a way to objectively compare safety ratings between different eras of cars? The standards for safety ratings change periodically, and great strides have been made in safety. I'm curious how much risk I'd remove by upgrading from a 2010 to a 2015 or 2018. Driver death data [1] mentioned below may be the best source, just curious if there are any others. Also, [2] is a really interesting crash test…

As much as I like driving a car into the ground, I think getting a new car every 10 years is the sweet spot between safety upgrades, frugality and shiny new toy. Here's some highlights and keep in mind there's a lag between the first car with a feature and it becoming a widely adopted standard. And of course each new feature is always being improved. 1959 - First car with 3 point seat belt as standard 1971 - First ca…

>As much as I like driving a car into the ground, I think getting a new car every 10 years is the sweet spot between safety upgrades, frugality and shiny new toy.

The balance between frugality and safety upgrades is my internal struggle. Growing up, I remember it being a big deal when cars hit 100,000 or 120,000 miles. That's pretty common now, which makes the safety aspect more important.

I average ~12,000 miles/year, which is pretty close to the US average of 13,500 miles/year. With cars lasting so much longer than they used to, buying an older low-mileage car means you'll be driving a pretty old car by the time it becomes high mileage.

I'm going to run my 2010 Civic for awhile longer, then probably look at a ~3 year old car with ~80,000 miles. I can run it for ~6 years, end up with ~150,000 miles, then sell it to a high school kid who is only going to put 5,000 miles/year on it.

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

#193

Post by hinkley » Tue, Jun 30, 2020, 7:10 PM UTC

ummonk wrote: post #156

Is passenger side small overlap really a common form of collision? I would imagine since everyone drives on the right side of the road, most higher speed small overlap collisions would be on the driver side. Perhaps car manufacturers were right not to optimize for passenger side small overlap collisions? Frankly though, this is an area where I don’t think it is all that harmful for the metric to become the target, si…

Passenger side small overlap seems like something that would happen when someone tries to pull into your lane. As you say, the average delta-V for that collision is going to be much lower than the driver's side collisions (passing on the right, or crossing the center line)

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

#194

Post by hinkley » Tue, Jun 30, 2020, 7:12 PM UTC

One of the more morbid yet profound aphorisms that I know is:

Safety regulations are written in blood.

FAA regulations are often based on finding a new way to break an airplane that they hadn't diagnosed before. Safety ratchets up as new failure modes are discovered. If cars keep showing up pancaked it a particular way, then guidelines change.

Until those failure modes are regulatory capture, at least.

In the 90's the collective wisdom said you could shock people into doing the right thing, and on several occasions they preserved a badly t-boned vehicle to illustrate why drunk driving was such a bad thing.

These days the pillars are huge and well-connected. I'm not sure you can crush a car that badly without a commercial vehicle. They test for that stuff.

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

#195

Post by Bartweiss » Tue, Jun 30, 2020, 7:14 PM UTC

im3w1l wrote: post #129

cybrjoe wrote: post #82

Earlier quoted context omitted.

Serious question, how is this different from what VW did? There was a lot of talk about complicit engineers vs. shady management in that discussion. Did this ever make it to production?

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 differently also. In the anecdote, the product being tested was the same one being sold, and there's no sign the heater was worsened to improve test performance. The designers just picked the best-scoring option among some reasonable configurations. (Frankly, once they noticed that issue, what were they supposed to do? Pick the worst-scoring, or pick the spec out of a hat?)

In the VW story, the test-bench vehicle was fundamentally different from the market vehicle, and the road version was designed to behave worse on the metrics to get other gains. I happen to know someone who bought a diesel Jetta specifically because it was more eco-friendly than other options, and I think he'd draw a clear line between tuning for test metrics and VW consciously lying to their buyers.

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

#196

Post by throwaway0a5e » Tue, Jun 30, 2020, 7:14 PM UTC

hansvm wrote: post #159

kcolford wrote: post #32

Earlier quoted context omitted.

I think it's more relevant how much each group drives. Everyone I know who drives a lot espouses how important 4WD/AWD is, whereas the people who barely drive at all aren't willing to pay so much more for it.

I've always found it a bit amusing how common this belief is, especially in snowier areas. Most people would be safer without 4WD/AWD -- less likely to oversteer around a curve, and less likely to drive aggressively because of a false belief that AWD will help you stop faster or have better control of your vehicle in low-traction conditions. It's still a really nice feature that has its uses, but most people I talk t…

Comments like this remind me of the Principal Skinner quote "surely I am right and it's everyone else who's wrong" or something like that.

Everyone(TM) prefers AWD for it's superior go-ing abilities even if the turning and stopping abilities are unchanged. An obnoxious vocal minority tells them they're wrong. Since it's a matter of performance and consensus the overwhelming majority can't be wrong though.

Anecdote time:

I have a fleet of several identical station wagons. There is one of them that's FWD. They all use the same brand of tires. It's as close to scientific experiment as you can get. The FWD one sucks in the wet/snow/anything but dry. Forcing all the horsepower to go through just two tires that also have to do double duty and steer the vehicle make it peel-out city. The AWD ones are much more tame. But the FWD one gets a couple more mpg so we keep it around. Given the choice between snow tires on the FWD wagon and driving one of the AWD wagons on all season tires I will pick the AWD one every time. Having a marginally shorter minimum stopping distance and being able to corner better is less of an improvement than having a substantially reduced likelihood of peeling out on an uphill.

The above opinion is shared by the other four people who have to drive the vehicles in question which is why I'm the one who gets stuck driving the FWD one when it snows.

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

#197

Post by bananamerica » Tue, Jun 30, 2020, 7:18 PM UTC

chickenpotpie wrote: post #102

ryanianian wrote: post #52

Earlier quoted context omitted.

Isn't this a symptom of the exact testing procedure being a white box that rarely changes? Many students would score 100% on the ACT if the entire question and answer set were widely published. Also that seems like a silly test. Couldn't you just measure the differential after mixing? TFA also mentions it, but noise is also a factor. For the water-heater example, the limited number of probes introduces a known amount…

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?

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

#198

Post by SkyBelow » Tue, Jun 30, 2020, 7:22 PM UTC

tialaramex wrote: post #138

SkyBelow wrote: post #127

Earlier quoted context omitted.

There is a difference between memorizing enough information to ace a test and sneaking in notes that aren't allowed to ace a test. And people would also say it is wrong to steal a copy of a test and then memorize the answers to it to ace a test. But what if a professor uses the same test every year (maybe changing a few numbers but in a way that only impacts the calculations, not the way to solve it) and people study…

If you cheat in most such tests it just means you miss out on actually learning what you were supposed to. If it wasn't your intention to learn anyway I guess that's fine. Rarely the purpose of tests is to assure the public of your fitness (e.g. a driving test) and cheating those might be a problem, but if you cheat my CS 101 course and then struggle because you needed remedial classes but the cheated test means you…

There are a few other issues with cheating, such as devaluing a degree for all others who didn't cheat to earn it.

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

#199

Post by sangfroid_bio » Tue, Jun 30, 2020, 7:25 PM UTC

wscott wrote: post #30

The article reminds me of a story from a friend that was an engineer at a major water heater company. Modern electric water heaters are already fairly efficient, but every model gets tested by the government to be given an efficiency rating. The test consists of replacing the anode with a rod that has a series of temperature probes to measure the water at different levels in the water heater. Since water would form l…

There was a scandal a few years back when a Kaggle team did something simar with ML. They were treating the competition as a black box and optimising for the unknown dataset instead of actually building better AI algorithms.

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

#200

Post by alistairSH » Tue, Jun 30, 2020, 7:26 PM UTC

henryfjordan wrote: post #85

cybrjoe wrote: post #82

Earlier quoted context omitted.

Serious question, how is this different from what VW did? There was a lot of talk about complicit engineers vs. shady management in that discussion. Did this ever make it to production?

VW actively detected if the car was on a treadmill testing setup and electronically changed the engine parameters to make it run cleaner. The water heater was designed with the test in mind but functions the same regardless. VW was clearly cheating whereas the water heater is taking advantage of known deficiencies in the test (is that cheating?).

I'm not sure that's an accurate description of VW's cheat.

VW knew the parameters of the test (stationary car, no steering input, prescribed throttle inputs, etc). And they configured the car to pass the test.

Water-heater engineer knew the parameters of the test (number and location of probes). And he configure the water-heater to pass the test.

Same-same. In both cases, an engineering team willfully committed fraud to improve their sales figures.

The only differences are a bit pedantic. VW's "configuration" was more elaborate. But, both groups gamed the test.