Susan Durham <sdurham@BIOLOGY.USU.EDU> wrote >Say you have a mechanical device that must withstand 20 lbs of pressure. >You test 10 of these devices with a machine that is able to deliver up to >400 lbs of pressure. None of the 10 devices fails. The final recorded >pressure measurements for these 10 devices are close (but not exactly equal) >to 400, with some variability but with a very small standard deviation. You >think you need a confidence interval for the pressure at which the devices >fail. > >This question was posed to me by a fellow shuttle passenger on the way home >from the airport. To an ecologist, I probably would say, "Wow, who needs >statistics on data like that?!" But this is a medical device so >expectations might well be different. I told him I'd ask around. > >My questions are: > >Even in a regulatory environment, is there any point in computing a >confidence interval on data of this nature? If so, how might this >confidence interval be computed? > >Is another form of assessment appropriate in this context (e.g., some >probability of failure)? > >I had one QC/reliability course many, many years ago and am essentially >clueless on this topic, so any suggestions would be welcomed. >

I don't know much about this field at all, but it seems to me that the relevant statistic is not a CI around the 400 pounds, but an estimate of what proportion would fail at 20 pounds.

Perhaps there are physical properties similar to a dose response curve that let you estimate this from the data you have, but I don't know.

Peter

Peter L. Flom, PhD Statistical Consultant www DOT peterflomconsulting DOT com