Bias correction factor for MSA Color study

Laura_Higgins

I am helping a customer who is trying to figure out how do make a correction factor from an EMP/MSA study. He is measuring color - he has a set of chips that don't change color  with temperature, humidity, etc. He has multiple locations, and some locations have more than one machine. He want to be able to tell if Red from Machine 1 and Machine 2 at location A are the same (so add some correction factor to the mean based on the bias). It's not a machine that I think he can change (spectrophotometer) so he wants a correction factor for his data.

I've attached an example with made up data (so forgive me Xan if you look at those color value and laugh, or anyone else that really knows color) as an example for him to think about his design. How would he come up with the correction factor? I think it is going to just be the +/- mean for each color nested within machine and company, but want someone else's opinion. Am I missing something here? Any other advice on doing this? Of course he won't have the same number of machines at teach location... He is creating a design now so this is the time to help as much as possible. Thank you!

Color EMP.jmp

Peter_Bartell

Laura: We used to deal with sort of problem all the time in my former role at Kodak where evaluation of color (typically in the printing industry....not paint chips or fabrics let's say) was something that was commonplace. Chances are good he's using either, as you suggest a spectrophotometer or colorimeter to create the L* a* b* values. I suggest he think about a different metric since he's mostly worried about some sort of correction or bias factor...there is an allied metric called delta E* ab which is described here:

Color difference - Wikipedia, the free encyclopedia

From here he can adopt any reference location as such, and then talk in delta E* ab space, not L* a* b* language. Delta E* ab is meant to correspond to the way humans perceive color differences. Generally speaking, the rule of thumb is humans can't for the most part, detect visually a difference between two samples if the delta E* ab < 2.0. So right there he shouldn't worry about from a practical perspective 'correcting' for anything less. But that's my two cents.

Give me a call next week and we can talk more if you would like. I think in general the structure of the experimental design is sound.

But there are all sorts of procedural gotchas measuring color in this way that can jump up and bite too which might influence bias. So standardizing the measurement system on ONE unifying method can sometimes reduce or even eliminate the need for these 'correction' factors. For example, if his materials are translucent, then a choice and type of sample backing material is critical. There may be sample handling issues involved too like orientation of the sample wrt to coatings and the illuminant. Or what are the specific operational characteristics of the spectrophotometers in question? For example, if he has one set up for what might be known as "D50 illuminant, with a 2 degree, standard observer" at one location, vs. "Something else" which is entirely possible (as we often found) between two locations...then all bets are off wrt to the data.

Here's just a little bit more about illuminants which is what I'm talking about above...if they aren't the same within his population of locations and spectrophotometers...then don't bother collecting data.

Standard illuminant - Wikipedia, the free encyclopedia