It isn’t logical to compare “Test Accuracy Ratio (TAR) with “Test Uncertainty Ratio” (TUR) when we start arguing about measuring stuff…

We often say that we want our measurements to be reliable and trustworthy. We may go further and invoke “TUR” or “TAR” in an effort to shore up this reliability. ?Even this simple sentence already contains some unspoken assumptions that could come back to bite us in the ass, especially if we decide to brawl over which ratio is "best".

?

For example we (or at the very least, I) assume that all measurements aren’t really measurements at all unless they are products of a system to which we periodically apply traceable standards to the measurement devices that we may use later to render these measurements. Without this underlying assumption of “Traceability”, everything turns to mud very quickly.

?

But returning to my title, is the statement that “all measurements have an attribute that we call accuracy” even roughly equivalent to the statement that “all measurements have an attribute that we call ?uncertainty” ?

?

As a matter of fact these two statements do not exist in the same "plane" and by that, I mean that there is no direct way to compare them because the two words have no equivalence of any kind at all. This in turn means that if we continue to insist on comparing “TAR” and “TUR”, we will next detect a logical minefield at our feet, if we and our clients are lucky…

?

Put simply, any statement of measurement accuracy that we may ever make has to have an uncertainty component attached to it whether or not we acknowledge its presence. We never see TARs deployed along with their stated uncertainty, but its always there!

In contrast, any statement of measurement uncertainty can stand on its own without needing any reference to “accuracy” of any kind. That is why we cannot pit these two concepts against each other. To go right ahead and do so anyway creates a “false dichotomy”.

?

To paraphrase Dr. Kristen Lennox [i], “uncertainty” is a technical term that we have, over the last 70 years, managed to define with increasingly exquisite formal and global precision. In contrast, “accuracy” is just an English word that we have grabbed and slammed into a convenient position, since it was already lying around in our vicinity.

?

To further add to the confusion, the words “accuracy” and “uncertainty”, as they apply to measurement functions come from completely different historical contexts. Using them as a basis for creating two measurement quality “rating” ratio systems is asking for trouble, and of course that is what we have gifted to ourselves. But, there is no other realistic path forward that anyone has suggested to us. So, play on!

?

Three or four hundred years ago, “accuracy” was the revolutionary measurement concept. Its appearance encouraged scientists (or those who would someday call themselves “scientists”) to figure out the rules that would apply to a group of measurements meant to characterize a single parameter. It was either that or follow the advice of one "thought leader" at the time who advised measurement producers to select their “best single guess” as to the correct value after making a series of repeated measurements.

This is the first article that I have written since coming back from a really nasty case of pneumonia. I am glad to report that this article just about wrote itself so it doesn't qualify in any way as a deathbed confession and like nearly all the others is Passive Voice Free! I hope that you liked it or at the least found it thought provoking.

?

[i] https://youtu.be/eDMGDhyDxuY? In this video Dr. Lennox discusses Bayesian and Frequentist statistical approaches. She is brilliant. Not only that, she is also funny, and that is practically a unique quality for anyone sporting a Ph D. in statistics while filming at their Lawrence Livermore Labs workplace! Stephen says check this out!