Question about Response Surface and Blocking

Question

Hi JMP Community,
I am interested in conducting RSM studies for two factors using CCD Orthogonal setup including 3 replicates. After finishing the setup, I get 52 as my total sample size. However, I won't be able to test all the samples in one day. I wanted to ask if it is possible to combine blocking with RSM. I also tried using CCD Orthogonal blocking, but it's only giving me two blocks since I am testing the factors at two levels - low and high. 
I have played with custom DOE designing. However, with custom DOE, I don't get similar levels that are bested tested in my RSM setup.

Any help would be greatly appreciated.

Thank you.

Victor_G · Answer

Hi @BetaAdjustment6,

Welcome in the Community !

Just some questions and clarifications from my side to continue the discussion :

You mentioned your DoE setup includes 3 replicates, for a total of 52runs. That means you have 13 runs per "whole design structure".

Can you run 13 runs in one day ? Or 26 ?

If yes, you could maybe split the runs by respecting the design structure and the replicates, and add a blocking factor for the replicate number (depending on the number of design structures you're able to run in one day, either a 2-level blocking factor if you're able to run 26runs in one day, or a 4-levels blocking factor if you're able to run 13 runs in one day).

There are also other options :

* Use Custom Designs and specify a RSM model with a blocking factor (and the number of runs per block/day you're able to run). But Custom design may only provide axial points with values -1 and 1 (so a Face-Centered Central Composite Design : https://www.itl.nist.gov/div898/handbook/pri/section3/pri3361.htm)
* You can also create your CCD Orthogonal design, and use the platform Custom Designs to introduce a blocking factor.* First, create the CCD design with replicates.
* Then, use the Custom Designs platform and select your covariates Factors to import your design table :

* 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".

This way, your CCD Orthogonal design with replicates could include a blocking factor to take into account possible day-to-day variation.

Hope these suggestions may help you,

statman · Answer

First, welcome to the community!

I don't understand what you mean by "won't be able to test all the samples in one day"?  I assume the samples are experimental units.  Are you concerned about the measurement process or measurement error?  Do you think the experimental units will change (e.g., degrade) with time?  Are you experimenting with factors in the measurement process?

I don't know the situation enough to provide specific advice, but here are my thoughts, in general.  Blocking is a strategy to handle noise factors in an experiment situation.  If you are doing response surface types of designs (e.g., optimization type designs), you are optimizing levels for factors already discovered to be of interest from previous study.  IMHO, you should already know the effects of noise BEFORE conducting RSM.  If you don't your inference space will be insufficient to use a more complex model provided via RSM.  Your process should be consistent and stable before optimization.