Design power

[#1]

Dear all, 

I am running a DoE experiment and trying to check the power of my design before starting the experiment. I followed exactly what is written in the JMP documentation, but I don't see the value of the power (screenshot attached)

I will appreciate your feedback 

[#2]

Hi Said,

It appears that you are following the example from JMP's DOE Guide - great!

As mentioned in the guide, "power is the ability of detecting an active effect of a given size" so, while you may be looking for an overall experimental power, the key is to take a look at the power provided for each term in the model you would like to build.

The Station variable is a 3-level categorical variable. JMP shows N-1 terms for categorical variables in the Power Analysis tab, hence why you only see Station 1 and Station 2. The overall power to detect the given difference between Stations is indeed the value you have circled in the attached .png (vs. the power to detect continuous factors which is shown above the 'Apply Changes...' button). There is an explanation for how to interpret these values at the end of the Power Analysis section in the DOE Guide.

[#3]

Hi @Jeff_Upton , 

Many thanks for your quick reply and all the information you provided. In my case, I am using the DSD (with 7 continuous factors) and trying to check my design's overall power. Still, I do not say it under the design evaluation -> power analysis. From my reading I know that we can evaluate the design according to its overall power (in my field they say that 80% is a good power level), however, I have no idea how to evaluate the design according to the power value for each term in the model?

Indeed is there a formula for the calculation of the overall power using the power attributed to each term in the model?

Many thanks in advance

[#4 · Accepted solution]

Hi @ELH,

Also great to hear you are experimenting with a DSD! Screening designs help one to understand which factors are most important in a process. You mention an 80% power level...this is a guideline for screening designs used across many fields. Experimenters want power for each factor main effect to be over 80%. So, in your initial screenshot, you would want the power for each term to be 0.8 or higher (only Station satisfies this criteria in your current design). This may be what your readings are referring to but I cannot say for certain.

You can improve power by increasing the number of runs or decreasing significance level. You could also decrease the expected RMSE to increase power but this would require prior measurement system optimization to reduce measurement error, for instance, which is not the focus of this DOE.