Best use of DOE Run Budget in Screening Design

[#1]

I'm looking for some guidance from the group. I am preparing an example to use in training with Custom Design. I want to demonstrate a screening design with 5 continuous factors. I have a run budget of 15 runs.

Using custom design main effects model with 12 runs (assuming default signal to noise assumptions) I get a power of 0.821

My question is about the best way to use the extra 3 runs

• Do I run centrepoints ?
• Do I run replicates?
• Do I do additional runs to fill out the design space?

I'm guessing it depends what my primary concerns are and the purpose of the screening DOE.

I am looking to develop some simple guidance for new DOE practitioners along the lines of   "if you are primarily concerned about better understanding the noise then run 3 centre points" or "if you are primarily concerned about .....then run 3 replicates "  or "if you are Primarily concerned about  understanding .....then do additional runs in different parts of the design space"

By the way I'm guessing the best answer might be to use the 15 runs and do a definitive screening design, but that is the topic of a separate module and I want to stick with custom design for this scenario

Many thanks for any thoughts

[#2]

Hi @ar2,

In your situation, I would simply run all these scenario using Design Explorer. You can then obtain a table with most of the design performance diagnostics, for different run sizes, replicate runs, centre points, optimality criterion, etc... It won't answer everything, but it's a good start, based on metrics and values.
You can see an example on how to compare different designs with the example I provided for DSDs here : Blocks and Center points for a definitive screening design 

The centre points are commonly used for three main purposes besides some design performance metrics :

1. Estimate pure error for the lack-of-fit test, to evaluate if you model seems adequate.
2. Decrease variance prediction in the centre of the experimental space.
3. Detect curvature in the response(s). But this curvature effect can't be linked to a specific factor quadratic effect.

So the choice of adding centre points (or adding quadratic effects in the model with estimability "Necessary" or "If Possible" if you have a low experimental budget) is mostly guided by the prior knowledge you have about your system, and any quadratic/curvature effects you may have for some of your factors.

The only situation you wouldn't cover with this platform is the augmentation of your design using Space Filling Augmentation. But in this last case, it's mostly a question of coverage of the design space than improving model-based design metrics.

Hope this answer can help you,

[#3]

Thanks Victor

Is Design Explorer part of standard JMP or is it in JMP pro ?

I haven't used it and I'll take a look. Bearing in mind that the trainees are totally new to DOE and JMP I have some concerns that Design Explorer might be a bit overwhelming for them. I was wanting  develop some very simple guidelines, but perhaps playing around with Design explorer I can construct these

[#4]

Hi @ar2,

Design Explorer is available for every JMP licences (standard and Pro).
You can use it to test multiple options and then extract some knowledge/information from your testing to make the conclusions clear for the students (using visualizations and simple analysis to display the pros/cons of different situations).

As student get more comfortable with DoE, I would encourage you (and them) to use this platform for concrete use cases/examples, as it may allow them to better compare, evaluate and choose the designs based on relevant metrics, and it could be a nice exercice for them to explain and justify their design choices with the support of this platform, based on their study objectives on specific use cases.

Best, 

[#5]

Thanks Victor - I'll take a detailed look at the platform