Date: Mon, 11 Sep 2000 10:31:52 +0100 From: Steve Ellison To: iso25@fasor.com Subject: Re: Proficiency testing RE14 Philip Stein wrote: >We assessors are often recommending that labs severely round off >any 'precision' they calculate in uncertainty estimates, since the >underlying science for most Type B's is usually only an educated >guess. Oho! So you thing the type A's are better, do you? Type B is stuff off calibration certificates; effective degrees of freedom up in the 30's on a good day. Type A, on the other hand, is the lab's observed precision. Degrees of freedom don't often get much above 10 in testing. With that number of degrees of freedom, a standard deviation is well and truly uncertain in the FIRST significant figure. A factor of two from one time to the next is 'normal'. Not much room to be throwing bricks at Type B! I can only excuse the slip ;-) if you're talking about 'professional judgement', which has an extremely bad press among statisticians when it comes to quantitative estimates of confidence. But even there, a decent measurement lab knows quite a bit about how bad it's 'normal' process can get; type B for known effects is unlikely to be noticeably worse than type A. However, your conclusions are sound. Quoting an uncertainty to more than one sig. fig. is probably being optimistic; quoting to three or more sig figs is just wasting ink!