DSD for 4 continuous factors

[#1]

I have four continuous factors. I want to optimized the process that's why I clicked add blocks  with center runs to estimate quadratic effects.

I click the script "Evaluate of Design" after making the table. I clicked RSM. The values of 2FI and Quadratic in Power Analysis are small. I also noticed that there is no quadratic term for my 4th factor.

[#2]

Hello @Alainmd02,

First of all, let's remind that definitive screening design is a screening design, so its aim is to detect significant main effects from a list of factors (although it is also capable of fitting some interactions and quadratic effects if remaining degrees of freedom allow it).

So in this sense, it's normal to have lower power for interactions and quadratic effects than for main effects, since the main effects in this design are not aliased with any 2-factors or quadratic effects. On the opposite, some aliases are still present between 2-factors interactions and quadratics effects (see correlation plot done on your example with 4 factors), so detecting precisely the significance for QE and 2FI will be more difficult with the presence of aliases (hence a lower power).

DSD is more recommended for 5+ factors; for less than 5 factors it may be easier and more efficient in terms of experiments number to use Custom Design.
 
Since it's a screenign design, its aim is not to fit a full RSM model (and JMP should have warned you when clicking on RSM that "This design cannot fit the specified model. Inestimable model terms have been removed"), so that's why you don't see quadratic effect for your 4th factor (lack of degree of freedom to precisely calculate this effect estimate).

In order to create an RSM design, you'll have to :

• go through augmentation of your DSD (go to "DoE", "Augment Design", and when augmenting, click on RSM in the model part, and JMP will recommend you to realize at least 4 more experiments, so total 22 experiments),
• or change your strategy :
  • create a Custom Design that fit RSM model (21 experiments recommended by JMP), or
  • use classical approach (Box-Behnken/Central Composite Design, from 27 to 36 experiments depending on the model chosen).

At the end, always compare the different models created (with different number of runs), to be able to choose the most interesting design, based on the best compromise between design performances (power for each effects, variance prediction, aliases...) and number of experiments.

Some ressources on DSD :

Using Definitive Screening Designs to Get More Information from Fewer Trials - JMP User Community

Definitive Screening Design - JMP User Community

I hope it will help you,

[#3]

Hello Victor,

Right now, I have now 4 continuous factors and 1 categorical factor.

Can you recommend DSD?

Why is it RSM in Custom Design have always low power for QE?

What minimum power for QE is enough to consider it as good design?

[#4 · Accepted solution]

Hello @Alainmd02,

If you already now that these 5 factors are of interest/significant for your study and that you want an RSM model, I would go directly to Custom Design (or classical RSM designs in the case of only continuous factors).

In this case DSD might not be the best choice, as the emphasis will be done on screening, and you will not end up with a complete RSM model just with one DSD (see my answer before; you'll also have lower power for QE than with a Custom RSM design).

There is no rule regarding the minimum power for QE, but you have to expect lower power for QE than for ME or 2FI. I would not be too anxious about lower power for QE, since the power for QE will also be dependent on how many factors and which factors enter the model. If you have remaining degrees of freedom, then power for QE might increase.
There was a post about power (Solved: Re: Should I consider power analysis in DOE? - JMP User Community), where @Phil_Kay answered and explained the use of power for evaluating designs. You might find it useful also in your case.

I hope it helps you,

PS: If you find an answer that solves your problem, don't hesitate to mark it as a solution to help other Community members track the solution to your question

[#5]

Hello @Victor_G,

Thank you for being generous, much appreciated. I think you have already solved my problems.