Subscribe Bookmark RSS Feed

Definitive screening Design with Skewed Center point

chemguy

Community Trekker

Joined:

Jul 26, 2013

Hi,

I have 7-factors and I am planing to apply a DSD. One of the factor needs to have a skewed center point of 30% (10%-30%-60%) due to experimental limitation instead of 35% required by DSD.

I created a DSD for 7 factors with normal default ranges and I created another DSD but manually changed the center point to desired 30% instead of 35%. Using "Design Evaluation" feature I compared both the design and surprisingly there is not a huge penalty for DSD with Skewed Center point for one of the factors (see attached outputs).

So my question is:

  1. Is this approach of manually changing the center point level in DSD acceptable
  2. Will the DSD with Skew Center point will still meet my objective of identifying main effects (without alias), fit quadratic effect (if any) and fit 2 or 3 highly significant interactions.
  3. Will there be any complications in further augmenting this design.

Looking forward for your input on this.

Thank you.

1 ACCEPTED SOLUTION

Accepted Solutions
Solution

Making such ad hoc changes to a design is what Ron Snee calls, "exercising the boss option." You did exactly the right thing by checking the design diagnostics using the Evaluate Design tool in JMP's DOE menu.

As you can see, your slightly off-center setting has a very minimal effect on the various measures. I would say that there is no practical difference.

The DSD you constructed will do everything that you want as long as your active effects are substantial compared to the error standard deviation.

4 REPLIES
louv

Staff

Joined:

Jun 23, 2011

I guess I would ask have you considered a 10, 30, 50 scenario? The diagnostics that you share with the skewing do not look bad but figured I ask.

Lou V

chemguy

Community Trekker

Joined:

Jul 26, 2013

Well both 10-30-50 scenario and  applying Custom Design is possible. I just wanted to discuss with Group about the pros and cons of playing with DSD designs like I did. Design Evaluation shows there is no loss but I also wanted to cross check my approach.

Solution

Making such ad hoc changes to a design is what Ron Snee calls, "exercising the boss option." You did exactly the right thing by checking the design diagnostics using the Evaluate Design tool in JMP's DOE menu.

As you can see, your slightly off-center setting has a very minimal effect on the various measures. I would say that there is no practical difference.

The DSD you constructed will do everything that you want as long as your active effects are substantial compared to the error standard deviation.

laural_

Community Trekker

Joined:

May 15, 2014

I recently ran into this problem myself.  In my case, I wanted to use the current product levels as the centerpoints, then vary the amount of chemicals around the centerpoints. Sometimes the high level was more than twice the total amount compared to the centerpoint level.  I didn't want to put in zero for any chemical, so I used asymmetrical levels for some chemicals.  Lou helped with the compare Designs to show it didn't make much of a difference.

Laura