Fractional factorials with categorical 2-level and 3-level factors

[#1]

Hello Dear All,

I have several DoE problems that include categorical
factors that I need to screen. The number of factors with two levels varies
between 3 and 5 while the number of factors with three levels varies between 2
and 3. It seems to me that I can use for all the situations described above a fractional
factorial design obtained from a L18 Hunter design which includes 8 two level
factors and 4 three level factors. The fractional designs are obtained by selecting an appropriate subset of columns from the original design,
as stated in the JMP DoE guide. If I use the JMP screening design platform to
design these screening designs JMP automatically shows a L18 Hunter design. My
questions are:

1. How does JMP obtain this design from the original L18 design? Which columns are
   deleted?
2. For my situations, are there better known designs with fewer runs to screen my
   factors? When I use the JMP custom design platform with a main effect model a
   12 run design is sometimes suggested, so I wonder if there is a known 12 run
   screening experiment (main effects only) for the situations described above or
   for some of them.

Many thanks!

Marcello.

[#2 · Accepted solution]

Marcello,

Not sure what version of JMP you are using but in JMP 12 you are able to use the Screening Design platform to generate and drive a 12 run design as you discuss by clicking the radial button specifying an orthogonal  or near orthogonal design rather than the catalog list of designs which gives the L18 Hunter.

[#3]

Louis, I am using JMP 12 and I tried your suggestion. Thanks, Marcello.

Da: LouV

Inviato: mercoledì 10 febbraio 2016 19.57

A: Marcello Fidaleo

Oggetto: Re: - Fractional factorials with categorical 2-level and 3-level factors

<https://community.jmp.com/?utm_source=JIVE&utm_medium=email&utm_campaign=System%20Email%20Tracking> JMP User Community

Fractional factorials with categorical 2-level and 3-level factors

reply from Lou Valente <https://community.jmp.com/people/LouV?et=watches.email.thread> in Discussions - View the full discussion <https://community.jmp.com/message/226697?et=watches.email.thread#226697>

[#4]

Marcello:

To pile onto what Lou has contributed above...in JMP version 12, as you suggest in point 2. in your original post, JMP's Custom Design platform can find a D-optimal design in 12 runs for your, what I'll call 'most saturated' factor/level scenario. See the factor specification window and design below. Power for effects with anticipated coefficients in the 1 to -1 range are pretty low...but it will work for a main effects only model. You ask if there is a 'known design' in point 2...I'm not sure what you mean by 'known design'...but this design is possible and for a main effects only model...with an eye towards further subsequent experimentation with key factors and levels...this just might meet your practical needs?

[#5]

Peter, thank you for your response, which is useful to me. I find advantageous the fact that I can obtain all the screening designs that match the number of 3-level factors and 2-level factors corresponding to my situations from just one L18 Hunter design. Marcello.

Da: [email protected]

Inviato: mercoledì 10 febbraio 2016 21.57

A: Marcello Fidaleo

Oggetto: Re: - Fractional factorials with categorical 2-level and 3-level factors

<https://community.jmp.com/?utm_source=JIVE&utm_medium=email&utm_campaign=System%20Email%20Tracking> JMP User Community

Fractional factorials with categorical 2-level and 3-level factors

reply from [email protected] <https://community.jmp.com/people/peter.bartell%40jmp.com?et=watches.email.thread> in Discussions - View the full discussion <https://community.jmp.com/message/226701?et=watches.email.thread#226701>

[#6]

fidaleom:

Your welcome. The beauty and benefits of JMP's Custom Design (JMP's deployment of the more generalized experimental planning strategy known as optimal DOE methods) platform is that it has many capabilities that allow the experimental planning team/individual to tailor a specific design to the practical problem at hand to include things like:

1. Disallowed treatment combinations.

2. Factor space restrictions.

3. Specific empirical model articulation BEFORE experimental execution so that the design will support the exact model of most interest prior to experimental execution.

4. Constraints on the number of runs...especially important if there are constraints such as budget, materials, time, where the # of runs is not a power of 2 if one's view of the DOE world is limited to the classic catalog of designs in the 2**k and 2**(k-p) fractional factorial design spaces.

5. And others...

[#7]

Thank you Peter. I definitely agree with you about the benefits of using optimal designs through JMP custom design platform. In fact in my case a 12 run experiment can be generated by using optimal designs, with a reduction of the number of trials from 18 to 12. Since I am collaborating with people that are not familiar with DoE methods, it is easier for me to refer to a design described in the literature (the L18 design) instead of generating optimal designs corresponding to the different situations at least as a first approach.

Da: [email protected]

Inviato: venerdì 12 febbraio 2016 14.30

A: Marcello Fidaleo

Oggetto: Re: - Fractional factorials with categorical 2-level and 3-level factors

<https://community.jmp.com/?utm_source=JIVE&utm_medium=email&utm_campaign=System%20Email%20Tracking> JMP User Community

Fractional factorials with categorical 2-level and 3-level factors

reply from [email protected] <https://community.jmp.com/people/peter.bartell%40jmp.com?et=watches.email.thread> in Discussions - View the full discussion <https://community.jmp.com/message/226734?et=watches.email.thread#226734>