How Do I Use DOE in JMP to Model the Effects of Process Inputs on Variance

[#1]

Greetings JMP Community,

My "quick" question is to find information on how to design and analyze a DOE in JMP used to model relationships between inputs and variance outputs?

Typical experimental designs are used to model the average output for a process to determine the significant inputs (screening designs) and possibly quantify the effects (response surfance models). I see a growing need to be able to model how model factors are related to a variance term created from several observations measured for each run. The sampling distribution differs as standard deviation typically follows a chi square instead of a normal distribution (used for means). The designs are likely resource intensive as the goal for the experimentation is typically to be able to quantify effects and optimize the model to minimize variance within runs. I assume that the non-linear DOE might be the way to go and I am interested in any tips on where to find more information and any examples.

[#2]

This should be a good place to start:

http://www.jmp.com/support/help/Loglinear_Variance_Models.shtml

"Note: The estimates are demanding in their need for a lot of well-designed, well-fitting data. You need more data to fit variances than you do means."

Regards,

Phil

[#3]

Thank you Phil.

I expect that the runs needed for even a small number of factors will not be trivial due to the requirement for replicates. I will take advantage of the link provided to put together some trial designs.