17149_en-US

liliconcom

fractional factorial design with 3 or more level continuous factors<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />

Hello,

a customer is asking if JMP allows to do fractional factorial designs with 3 or more level continuous factors (no component, with component (different levels). Or alternatively,

is it possible to use the fractional design it generates in the full factorial design set up and will it still be able to generate reliable data?

Thanks a lot,

Valerie

 

Re: fractional factorial design with 3 or more level continuous factors

I would agree with Malcolm's thoughts on getting a sense of those higher order effects.

One thing that tends to come up along with this is that the values in the middle (whether continuous or discrete numeric) won't show up as often.

An exception is the main effects screening designs from the screening design platform: it uses the values of discrete numeric factors equally often, but you would need to check design evaluation to see about higher order terms.

Re: fractional factorial design with 3 or more level continuous factors

The key word I'm going to focus in here in your request is 'reliable data'. The customer may want to spend some time using the Design Evaluation tools BEFORE running the experiment to compare different aspects of the design's power, aliasing, and so forth. Melded with their process knowledge this sort of pre experimental execution work can help them have the highest degree of confidence that their specific experimental goals and objectives have a 'reliable' chance of being achieved. So Malcolm's ideas are spot on to generate candidate designs...but I'd sure spend some time evaluating various candidate designs.

Re: fractional factorial design with 3 or more level continuous factors

Also if they have a large number of factors it will be costly in terms of total number of experimental runs to model many higher order effects. So best to also ask them what is the cost of an experimental run and if they have any limitations on their experimental budget.

Re: fractional factorial design with 3 or more level continuous factors

Hi Valerie,

Can you please ask your customer how many factors they want to investigate and the nature of the effects they want to model on the response - linear, quadratic, 2-factor interaction or higher order effects? We can then guide them to an appropriate design, which if they have a large number of factors will be DSD or Optimal Design. Both will achieve what I suspect is behind their intention to go beyond modelling linear effects for a large number of factors.

Malcolm