Definitive Screening design runs

[#1]

Hi

From the DSD paper it is said that the number of runs that I have to do is 2M + 1 (Even Factor) and 2M+3 (Odd Factor), but when i use the DSD window in JMP 14 any reason that my 4 factors actually shows that I have to run 17 runs instead of only 9?

Rgrds

Irfan

[#2]

There are a couple of aspects that determine the number of runs in the DSD now. First, later research in these designs showed that the power of the tests associated with these screening designs significantly increased when at least 4 extra runs are included, so that change is automatically in effect. You may change the number of extra runs at any time, though, including none. Second, a screening design is an advanced design for a specific purpose that might work only if the key assumptions hold. The sparsity of effects principle is one of the key assumptions.

So if you have less than five factors, JMP designs the DSD for 5 factors. (Essentially that behavior is saying that only 4 factors is not really a screening case so I need to protect you if you want to use a DSD.) That means the minimum number of runs is 2*5+3 = 13. Then the 4 extra runs gets you to 17 runs.

See Help > Books > Design of Experiments. There is an entire chapter devoted to the latest implementation of DSD in your version of JMP.

[#3]

Hi Mark,

Thanks

Rgrds

Irfan

[#4]

Hi Mark,

Are you able to elaborate which extra runs condition that the system auto generate? Is there a specific formula that the system follows or its in random?

Rgrds

Irfan

[#5 · Accepted solution]

The approach is simple. Your formula for the number of runs is correct. You increase the number of factors to get the number of runs, then throw away the 'fake factors.' Let's say that I have 8 continuous factors. The minimum DSD will be 17 runs. I want 4 extra runs for sufficient power. The algorithm creates the DSD for 10 factors, and then removes the last 2 columns in the design. The resulting design still possesses all the desirable  characteristics of a DSD.

I hope that I understood your question.

[#6]

Hi Mark,

Ok understand, thanks for the explanation

Rgrds

irfan

[#7]

Dear Mark, Thank you for your helpful comments. I had 2 to 3 factors (e.g., pH and Temperature) and wanted to have a minimum number of runs (max 10 runs) to identify meaningful factor (linear effect) and non-linear effects (e.g., interaction and quadratic effects). From your advice and what I understood, It would not be practical to choose DSD with JMP for this purpose. Can you please let me know what design you suggest me to use for this purpose? Also, when we run DSD, the software asks whether you want to do regular randomization or randomization within blocks. Which one would be more accurate? I appreciate your support.

[#8]

I suggest DOE > Custom Design for this experiment.

Blocking is used when there is an identifiable source of variation that is not one of the factors of interest but will vary during the experimental runs. It might be equipment, such as using 3 different ovens, raw material, such as using more than one batch, or time, such as many days. Each of these examples is identifiable as a source of variation. Blocking causes the design to minimize the correlation between the variation caused by these sources and the variation caused by varying the factor levels.