“True value” is a concept that is a lot like “Newtonian Mechanics”; not so much wrong as superseded.

I have just figuratively torn up about my fifth draft of a post about some of the recent efforts to re-write the GUM (ISO Guide to Estimating Measurement Uncertainty). I first found inspiration for this topic when I saw a draft of a paper by Hening Huang which CalLab magazine[i]?just published.

Huang wants to replace both frequentist and Bayesian statistical approaches featured in the GUM with his own improved version. I found his extensive use of metrology jargon as well as his high-level math intimidating, but I was curious enough to want to learn more about his positions[ii].

Despite my discomfort, it’s quite easy for me to identify a piece of both frequentist and Bayesian versions that Huang preserves in his version that I think should make us immediately put the brakes on his proposal. Its not his arguments, it’s his assumptions.

It is going to take a tremendous effort to change the course that the international metrology community has set for themselves through the years of work on various versions of the GUM starting in 1993. It would be a waste of time to even begin this investment if the new version preserves the concept of “True Value” and its subsidiary idea, “measurement error”[iii]. If we cannot achieve such a comprehensive improvement on the current path of the GUM approach, then I think that the rest of the world will continue to squeak by just fine until we can, thank you. To simplify things, one strength that Huang claims for his new version of the GUM is that it too, continues to preserve the old idea of “True Value”, and carry it into the future.

Since the concept of “error” is a logical and direct outcome of the belief in the existence of a “True Value”, let’s concentrate on dealing with “True Value”, and let “errors” fall away on their own. Such a perfect result as a “True Value” can only come from a perfect measurement. Should this occur, it would fly right in the face not only of all metrology but also statistics itself.

Let me stop for a moment and recommend the very best thing that I have ever seen on the 100 year-old topic of “Bayes versus frequentism”[iv].?In this superb video, Kristin Lennox says that “statisticians use probability to describe uncertainty”. If we buy that definition, and I certainly do, then we have to stop including “true values” or perfect measurement as concepts that could ever have a part of any discussion that involves statistical analysis and Metrology. I say that because if the rest of our approach has formal, mathematical support, the act of tossing in this concept stands out as a glaring exception.??“True Value’s” exists depends only on it being the only possible place to hide from uncertainty if we are determined to hide from it. A “true value” make all statistics superfluous.

If its true that “True Value” is a cherished belief held in common by Bayesian and frequentist statistics, then I say that this provides all of us with a great opportunity to at least see that this very old belief is?partly correct but still clinging to an element that will hold us back?if we insist on preserving it further.?

Let’s try the analogical approach…

This switch seems perfectly justified to me as a writer because this is the space within which “True Value” has always existed. “True Values” do not exist here on Earth, they are lofty ideals. Speaking of analogies: there is already a boat tied up at the dock into which we can and should transship the entire “True Value” package. That boat already contains Newtonian Mechanics which has also survived for about the same length of time. Even today, Newtonian Mechanics is still not “wrong” and still quite adequate and useful for dealing with some problems. Just don’t try using it to describe the orbit of Mercury or why GPS works on your phone- you’ll break it, and what possible useful point is there in doing that?

We have woven “True Value” into our core assumptions about measurement and very deeply, too. It’s a legacy from our past. If a guy as smart as Hening Huang says that we cannot possibly do without it, going forward, I consider that that statement is a gift from above. It reveals so clearly that we are discussing a sacred assumption that weakens us while helping link us with our past. One argument that Huang uses for preserving “True Value” is that our equations depend on it[v]. I say, write some new equations! If “True Value” is so profoundly central that its inclusion is necessary for the legitimacy of any and all formulae for estimating measurement quality, where is the proof to support that increasingly extraordinary claim? The rest of the science here in the GUM is changing rapidly, why do we cling to this undefended belief today?

Only when we let go of this legendary object can we stop hugging the shore and head out toward uncertainty.?This is my agenda. I tore up my previous drafts because I couldn’t state that position clearly enough to distinguish myself from Huang or anyone else afraid to look at the assumptions that they preserve when they support “True Value”.

Its far more difficult to identify and re-test our old assumptions than it is to come up with bold, fresh, new ideas. Meanwhile, the consumers of our measurement expertise are going through fundamental gymnastics, as well as their final consumers. What proposals can we “metrology practitioners” make that are actually useful to them?

Time to put up or shut up, what is my idea for replacing “True Value”?

I’ve been talking smack about any further use of “True Value”. What do I propose to actually improve things? Every time we might have used “True Value” let’s substitute the phrase “most likely value”. These three words appeal to me because this phrase immediately refers to the risk of any and all attempts to assign a value to a target. All measurement entails risk which happens to be far more concrete that “Truth”! The phrase “most likely value” leaves no doubt that the question we face is not philosophical, but statistical. Whenever we settle for a single representation of a measurement value, we immediately confront the risk that we are undertaking. Is that one single measurement attempt adequate to contain our risk??More globally?will accepting that single value contain the risk to the process lurking in the background? Sometimes it will, and sometimes it won’t. We own that answer without deflecting attention to Philosophy.

In summary

Employing the concept of a “True Value” in the midst of a technical approach (The GUM) that has required decades for us and our forebears to assemble, is simply a very surprising example of laziness on our part. Since the tide is so clearly against preserving this very ancient idea, can we please just cut to the chase?

[i]?https://www.callabmag.com/practitioners-perspective-on-the-gum-revision-part-i-two-key-problems-and-solutions/

?Thanks to Sita Schwartz, Editor, for this link!?

[ii]?A large majority of Huang’s citations in this paper remained behind a paywall for me that, as a retired person, I cannot afford to breech. Several of his citations sell for $20-30. It is a strange irony that “peer review” on this very modern topic come from an academic setting that connects directly to the Middle Ages. Traditions like tenure, passive voice, “publish or perish” are rather old. And if you happen to have a certificate of graduation from just about any “institution of higher learning”, I bet that there is some Latin on it.

[iii]?Please don’t get me started on the concept of “calibration error”, the bane of my existence for a decade in which I did field calibrations in Pharma!

[iv]??https://youtu.be/eDMGDhyDxuY?Will you commit ten minutes to this video? If you do, I can just about guarantee that you will stay until the end in which she reveals whether she is a “frequentist” or a “Bayesian”.

[v]?Hening Huang, Comparison of three approaches for computing measurement uncertainties,?www.elsevier.com/locate/measurement, 2010.