Print

Print


Nice to hear from another Canadian (Pak C Chan) on this topic. As I see
it, the problem with using the terms TE and MU interchangeability is
that this implies that both concepts are interchangeable, and this is
not the case. It is true that both concepts provide for a way of
quantifying the range into which a measured result will be with a
specified probability, but the two concepts differ dramatically in how
this may be done. Using different terms makes it clear which concept you
are using; I believe both concepts have validity depending on the
application.
The TE concept deals exclusively with Type A error, whether it estimated
from a method evaluation study or quality control data. On the other
hand, the MU concept includes both Type A and Type B errors, as long as
the error can be quantified in some way. This makes the MU concept a
powerful tool when the objective is to reduce error in processes by
isolating the contributing sources of error. Admittedly, in the examples
presented, Type B error was ignored for the sake of simplicity in
discussion.
As well, the MU concept excludes bias in the estimation of uncertainty,
although as PC  mentions there are suggestions in the literature about
including bias (though I think this is contrary to the what experts
advise) in estimating measurement uncertainty. The issue with bias is
that bias is unidirectional and can not be propagated in a
unidirectional way along with bidirectional error like imprecision, as
far as I know. If I am mistaken about this, I would be very interested
in hearing from anyone about how to include bias in the propagation of
uncertainty. 
I agree that TEa was originally used to designate a performance standard
and was not intended to be a performance characteristic. Nowadays, the
term performance standard is preferred for this purpose. However, the
way that it was being used by Andy Minett in a previous posting is
clearly as a performance characteristic and I must say it is quite an
elegant way to characterize potential error (TEa = CV(SDcrit + 1.65) +
bias). I still feel that it is important to distinguish between TE and
TEa when used in this way, although I accept that it is mostly used to
refer to a performance standard.
The frustration PC expressed about dealing with bias does not in turn,
provide justification for including bias in the estimation of
measurement uncertainty either. That is, if it is felt that measured
bias does not adequately reflect the “right” bias of reported results,
then what would be gained by including bias in the estimation of
uncertainty. In addition, if one truly believes that nothing can be done
about bias then why bother measuring it in the first place.
As a newcomer to this topic, I have no hesitation in declaring my lack
of expertise in this area. I welcome all feedback that would correct any
misguided thinking on my part. 
Regards to PC.

Cheers
David Parry
Winnipeg, Canada

This email and/or any documents in this transmission is intended for the
addressee(s) only and may contain legally privileged or confidential information.  Any unauthorized use, disclosure, distribution, copying or dissemination is strictly prohibited.  If you receive this transmission in error, please notify the sender immediately and return the original.

Ce courriel et tout document dans cette transmission est destiné à la personne ou aux personnes à qui il est adressé. Il peut contenir des informations privilégiées ou confidentielles. Toute utilisation, divulgation, distribution, copie, ou diffusion non autorisée est strictement défendue. Si vous n'êtes pas le destinataire de ce message, veuillez en informer l'expéditeur immédiatement et lui remettre l'original.