Date: Thu, 22 Apr 1999 16:04:54 -0400 From: "Matthew DeBell" <mdnopsamnet> Subject: Re: Caveat on death rates as an indicator of safety
sd wrote in message <371e91a1nopsam1.us.ibm.net>... > >[2 paragraphs omitted] > >While I'm taking the devil's advocate position here, it is mainly to point >out that while there can be many statistical confounders in a study like >that, the best information about how safe a car is remains the accident >reports which come in off the streets. They should not be taken as the >gospel, i.e. "Well, the Saab 900 got the best safety rating from the X >agency, so I can drive however I want without fear of getting killed in a >crash," but are better than the, "I love my Honda and it's the best car in >the world," yarns which contain little factual information at all. I agree with you about the accident reports, except I'm not sure that they offer the *best* data available. Possibly the crash tests are more valuable. I just don't know, because I don't know how reasonable it is to generalize from the crash test circumstances to actual accident circumstances. Obviously real crashes can be quite different from lab conditions, but I don't know how reliably we can expect crash performance in the real world to be consistent with lab observations. My non-engineer's intuition suggests it could go either way, depending on how the cars are designed. They could be optimized for all-around safety, in which case the frontal test might be very useful, or they could be optimized for the frontal test, in which case the frontal test is much less useful. Surely some designers do optimize for the frontal test so their product will look good, which is unfortunate. What I do know is that, given complex problems like this, the best way to go is usually to make multiple measures using multiple methods. Thus, we should take the accident data and the crash test results together with what information we have about drivers and use habits to make an educated guess about the safety of a particular model. If a car has both low fatality rates and good performance in tests, then I'm comfortable calling it a safe car. If the fatality rates are low but the crash test results are dismal, then I'd start probing for variables other than the car's design to explain the low fatality rates. -- Matthew DeBell mdnopsamnet
![]() |
![]() |
![]() |
![]() |
![]() |