DOE - blocking augmented design

[#1]

Hi,

I am planning a DOE to save on time/resources, no surprise here. I don’t have much experience with the topic, but would like to push for DOE instead of OFAT within the org I work at, I guess common idea too.

About the problem, there are 10 continuous factor, 3 responses. The experiments can be done at 4 labs at physically different locations. Lab 1 has more resources than the other 3, thus the preference is to run more at this lab. Each run takes 1 day.

My theoretical idea was to organize the experiment as below:

1. Custom design – “alias” optimality. Try to keep # of runs under control, scan for main effects only, run all at Lab1
2. Augment design from 1 – “D” optimality. All main effects & interactions. Obviously more runs thus this attempt to utilize all labs. Table below shows how the design would look like. Augmented design makes blocks for L1, Labs would be blocks for L2 .Level L1: Block B1 – Custom “alias” design, B2 – Augmented design; L2 – B1/Lab1, B2-B4 – other Labs.

Again, the idea is to run most of experiments at the Lab1 with other labs helping with the workload.

Questions:

1. I couldn’t find option to Block extra runs in augmented design in JMP. Is this not statistically valid approach? Or is it not available via GUI but still can be custom coded?
2. Alternative would be to run 2 separate DOEs, one to screen for active effects, other to test for main & interactions. This would obviously require more runs and is harder to justify. I understand a lot of info is missing here making it hard to give recommendations, but are there any other DOE approaches that would be “cost friendly”?

Thanks for the help.

[#2]

Hi @beluga111, 

Welcome in the Community !

Some questions and comments from my side before exploring the options/solutions :

• How much runs are planned for the first DoE ?
  As each run takes 1 day, I would also use Blocking in this first stage, to make sure you can assess a possible variation from week-to-week.
• Why are the runs for the first DoE only shared to Lab 1 ?
  I would use Blocking and distribute runs among labs in the screening stage, as you might be missing some active effects and not account and estimate labs variation if you only run the first screening experiments in Lab1.
• Why do you choose an Alias optimality criterion in the first DoE ?
  If your goal is to screen and identify main effects, I would try A- (or D-) optimality criterion, which is the recommended Optimality Criterion if you build your screening design with the Custom design platform. You could also use the Design Explorer to try and compare several designs in paralell (with different runs size, optimality criterion, replicates, etc...) and choose the most effective/adapted one.

Concerning your questions :

1. You can create another blocking for the "augmented design part", but this is not straightforward.
   First, create the augmented design, and create a new table with the new runs only using Create a Subset Data Table.
   Then, on the subset table, use the Custom design platform and select your covariates Factors to import your design :
   

   You can then add a blocking factor in the "Factors" panel for the labs, check the option "Include all selected covariate rows in the design", and "Make design".
   Your new "augmented design part" has now a blocking factor (you can change its Design Role to "Random Block" if needed), and you can combine your original screening design table with this new augmented design table to get the final expected design with two levels of blocking.

2. As you mention, creating 2 separate DoE could augment the required number of runs, and you wouldn't be able to leverage completely and efficiently the information gathered in the screening phase to create your second design.
   I think your original design idea is interesting, I would simply in the first screening stage use all labs and create a blocking factor for labs (as this would also give you important information about reproducibility early in your project), and augment this design on identified active factors to explore further interactions and quadratic effects. 

Hope this answer will help you,