Design of experiment for loglinear variance model

yuichi_katsumur

Hi,

 

Customer asked me about design of experiment.

Does anyone know how to design experiment when you want to fit loglinear variance model in JMP?

 

In documentation,

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.

 

Which platform is suitable when design experiment (ex. Custom Design, Non linear Design)?

How many replications is needed or recommended?

 

Thanks,

Yuichi

DonMcCormack

Yuichi,

A loglinear variance model uses both the mean response and variance in the analysis. You could use the Custom Designer or any of the classical designs to design the experiment. The difficulty is knowing how many observations will give you a good estimate of the variance. Assuming you could make some assumptions about the mean and variance of the response relative to design factors, you do this using the Simulate option in JMP Pro. Another approach would be to create the design as you would normally and use the Fit Loglinear Variance to see if you get a better fitting model. This would be similar to performing a variance stabilizing transformation.

Don