I believe there should be some discomfort or questioning that happens when choosing anything other than complete randomization in MSA. It is much easier to make operators blind to the run order if it is completely randomized. The risk of an operator biasing MSA results is lower when an operator is blind to the run order. I have many examples from personal experience of how experiments can be compromised when operators know what they are measuring. I participated in several MSA's and DOE's where a lab manager took the random run order and organized it into the easiest to manage and least random run order. I understand the desire to simplify the run order but I believe it should happen as a last result and only when it is determined that the test cannot be run in any other way.
My proposal
Make complete randomization the default for MSA design followed by batch repeat then fast repeat. Hopefully this will create enough cognitive dissonance when choosing the 2nd or 3rd options that the user will think about what they are losing. Perhaps there could also be something in the documentation that encourages practitioners to look at other designs that help deal with randomization issues and in all cases encourages making operators blind to the run order.
J
