A rose by any other name would smell as sweet…

is a quo­ta­tion by William Shake­speare from his play Romeo and Juliet meant to say that the names of things do not mat­ter, only what things are. In the play Romeo and Juliet, the line is said by Juliet in ref­er­ence to Romeo’s house, Mon­tague which would imply that his name means noth­ing and they should be together.

Does nomen­cla­ture really matter?

Why do we lawyers have prob­lems con­nect­ing and talk­ing with you scientists?

Why do you sci­en­tists seem so obtuse and need­lessly pedan­tic to us?

In part, it may be a nomen­cla­ture issue. I sug­gest that if we are going to try to open lab­o­ra­to­ries and make them be trans­par­ent and for lawyers and the judi­ciary to exam­ine their processes to ensure against unjust con­vic­tions, then we need to mind our nomen­cla­ture. I sug­gest that we use either the Inter­na­tional Con­fer­ence on Har­mo­niza­tion (ICH) and The Inter­na­tional Union of Pure and Applied Chem­istry (IUPAC) Gold Book definitions.

No place in foren­sic test­ing is this need to use the cor­rect nomen­cla­ture more impor­tant than when we dis­cus the valid­ity of a given method when it comes to any sort of test­ing. A small com­po­nent of valid­ity is metrol­ogy. We have dis­cussed metrol­ogy before here at the www.TheTruthAboutForensicScience.com blog as well as www.PADUIBlog.com as well.

The goal of all mea­sure­ment is to try to cap­ture the true value or the actual value of that which we are mea­sur­ing. How­ever, we can never, never do so. We can only attempt to design a method of mea­sure­ment where we have set up a process where we have deter­mined what level of risk we are will­ing to accept that we are wrong. Mea­sure­ment is the study of accept­able risk. What level of risk is accept­able that we could be wrong in our mea­sure? You see we are always wrong. It is a ques­tion of how much are we will­ing to risk that we are wrong and how wrong are we will­ing to be. Uncer­tainty Mea­sure­ment (UM), if prop­erly done, is the imper­fect embod­i­ment of the expres­sion of that risk.

You write as to “accu­racy.” Accu­racy (strictly in a ICH and IUPAC way) is a par­tic­u­lar type of assess­ment of a mea­sure­ment. Accu­racy is more prop­erly known as “bias.” Bias is the mea­sure of how closely the results are to the true value. It is char­ac­ter­ized by per­haps a high Stan­dard Devi­a­tion, but may or may not have a low aver­age devi­a­tion from the true (actual) value.

Then there is pre­ci­sion. Pre­ci­sion is an entirely dif­fer­ent type of ani­mal. They are inter-related and depen­dent vari­ables, but they are entirely dif­fer­ent con­cepts. Pre­ci­sion is more prop­erly known as “cal­i­bra­tion.” Pre­ci­sion is best defined as a mea­sure of how closely the results can be to one another. It is char­ac­ter­ized by a low Stan­dard Devi­a­tion, but may or may not have a high aver­age devi­a­tion from the true (actual) value. Pre­ci­sion is made up of repeata­bil­ity, inter­me­di­ate pre­ci­sion, and repro­ducibil­ity. Repeata­bil­ity is char­ac­ter­ized as the abil­ity day-in and day-out using the test, using the same method on the same instru­men­ta­tion on the same unknown arrives at the same result. Inter­me­di­ate pre­ci­sion is an expres­sion of with-in lab­o­ra­tory vari­a­tion: dif­fer­ent days, dif­fer­ent ana­lyst, etc. Repro­ducibil­ity is defined as the abil­ity of a test or exper­i­ment to be accu­rately repro­duced, or repli­cated, by some­one else work­ing inde­pen­dently. Pre­ci­sion should be inves­ti­gated using homo­ge­neous, authen­tic sam­ples over the long term.

There are three graph­i­cal rep­re­sen­ta­tions that best and most sim­ply show these concepts.

(This graph­i­cal rep­re­sen­ta­tion is the best one to describe these inter­sect­ing and depen­dent fea­tures as to a sin­gle mea­sur­ing event)

(This graph­i­cal rep­re­sen­ta­tion is the best one to show the inter­sect of these depen­dent fea­tures over mul­ti­ple mea­sures. As we can see from this depic­tion the goal of min­i­miz­ing risk of bias and cal­i­bra­tion error is a mov­ing tar­get. You adjust one and the other may suf­fer. It is also quite costly to min­i­mize both simul­ta­ne­ously. It is expo­nen­tially eas­ier and cheaper to cor­rect for cal­i­bra­tion error than bias error.)

(This graph­i­cal rep­re­sen­ta­tion is from Ted Vosk’s pre­sen­ta­tion at the AAFS meet­ing. I am unsure as to where he got it. This graph­i­cal rep­re­sen­ta­tion again looks at an indi­vid­ual mea­sure and shows the dif­fer­ence between Type I error and Type II error. Type I error can be termed, by and large, as a func­tion of bias; whereas, Type II error is, by and large, a func­tion of calibration.)