blocking in jmp

[#1]

hi,

i have a question about blocking in DoE. should we always follow the blocking option that JMP suggest with each design?? or can we do it our way?? for example. i am looking to do Box-Behenken design which has no blocking option at all but i was thinking of (doing all 13 run) in one day as the experiment is very short. then reopeat this over three to five dys and use day as a block were the in total i have 39 run were each 13 run goes for a day. am i correct in my thinking?? 

Regards

[#2]

Hi @NourNashed,

Can you provide more details about your experimental setup, objective and design ? I can't find a default 13-runs Box-Behnken design option for 3 continuous factors (minimum is 15 including 3 centre points). Did you change the default number of centre points from 3 to 1 to obtain 13 runs ?

If yes, there are two options to get 39 runs with 2 replicates and the use of blocking:

1. Create your 39-runs Box-Behnken design with 2 replicates :
   

   Then, go to the Custom design platform and use these runs as Candidate set in the Custom Design platform (simply select your factors in "Select Covariate Factors"), add a blocking factor in the "Factors" panel (with the specification of 13 runs per block), and specify a RSM model in the model panel). Check the option "Include all selected covariate rows in the design" to make sure that all previous Box-Behnken replicated runs will be present in the final design with 3 blocks.

2. Or simply create the original 13-runs BB, and then use the platform Augment design to replicate the design 2 more times (number of times to perform each run = 3) and add a block column after the replication with the relevant column properties : Value Order (for the ordering of the blocks), RunsPerBlock (=13), Design Role (= Blocking) and Factor Changes (Easy), and increment the block factor value every 13 runs. The problem with this option is that the order of the runs is the same for each block/day, so it might be better to randomize within each block/day at the end to avoid any pattern/bias throughout the day that could impact the results (linear or complex time trend during the day).

Hope this answer will help you,

[#3]

As there is not enough information on the situation, I can only lend some general advice.  For industrial experiments (vs agricultural), I would only recommend 2 blocks.  The purpose of blocking is to:

1. Increase the inference space (this is done more aggressively if you purposely exaggerate the noise between blocks, like build level setting for factors).

2. Increase the precision of the design.  This is because the noise confounded with the block is partitioned from design structure.

The only reason to run more than 2 levels in industrial experiments is to add non-linear terms to the polynomial.  This is non-sensical for blocks as they are "sets" of noise, not a continuous variable (there is no such thing as a 1.5 block).

I will also add, I believe JMP "likes" to consider blocks as random effects (you can tell by the model created by JMP when using JMP to create the blocks).  If you have identified the noise confounded with the block, blocks can be even more beneficial as they can be treated as fixed effects.  This allows for estimation of block-by-factor effects (the absence of block-by-factor effects is a quantifiable definition of robust).  To do this, I create the blocks myself (both Complete and incomplete).

see Sanders, D., Leitnaker, M. G., & McLean, R. A. (2002). Randomized Complete Block Designs in Industrial Studies. Quality Engineering, 14(1), 1–8. https://doi.org/10.1081/QEN-100106880