Date: Fri, 16 Apr 1999 00:30:11 -0700 From: "Matthew DeBell" <mdnopsamnet> Subject: Re: Caveat on death rates as an indicator of safety
Eoin wrote in message <371673ef.32896823nopsam1.ibm.net>... >... >> (The sample sizes are not large enough >>to offer much confidence that the Dodge would still be higher than the >>Plymouth in a much larger sample.) > >Defend that statement. An ANOVA would be nice. Ha! An ANOVA. You are the first person on Usenet I've ever seen ask for an ANOVA. Actually, IIHS presents the relevant data on the page I cited in my previous post: http://www.hwysafety.org/crash/crashimg/ddrchart.pdf There were only 43 deaths in the Dodge and 25 in the Plymouth during the years studied. These numbers, divided by the number of registered vehicles and the average fatality rates, produce the respective fatality rates for these cars: 187% and 119% of average. IIHS reports the respective 95% confidence intervals for the fatality rates as 129-263 and 70-188. For the non-statistically inclined (assuming y'all even read this far), overlapping confidence intervals like these tell you not to bank on one figure really being higher than the other. Like when a public opinion poll shows that 43% of those polled support some measure and 40% oppose, but the margin of error is + or - 5%, the poll doesn't actually demonstrate that those in favor outnumber those opposed in the real population. With the cars here, there haven't been enough tests -- i.e. enough registered vehicle years and fatal crashes -- to make reliable predictions about the probability that any given Neon driver will have a fatal accident. The appearance that the Dodge is much different than the Plymouth may be deceptive; it's distinctly possible that the difference is just a statistical fluke. -- Matthew DeBell mdnopsamnet
 |
 |
 |
 |
 |
