Interpretation of Power
The power of a design is the ability of the design to detect that specific terms are statistically significant - or the ability to find significant effects. Look at the output below for an example.
The output says that this design has a 44.6% chance of identifying a 1
standard deviation (Signal/Noise ratio) difference between the largest and
smallest average response with respect to the levels within A. Meanwhile there
is a 5% chance any effects found as significant only occurred due to random
chance.
Because A has the smallest power, it should be used to determine the power for
this particular design. A has the smaller power because it has the most
levels.
For practical use of this information, think as follows:
How much does the response need to change in order for you (or your customer) to care about it? That is the effect you want to detect.
What is the normal process variation for this response? Typically there is some historical data on responses that can be used to estimate a standard deviation.
A rule of thumb is that the power, which is the probability of detecting an effect of the given size, should be approximately 80% or larger to be confident in finding the desired effects.
Power can be determined during a factorial build, as well as under the evaluation node - the evaluation node has a different format for the display, but power is interpreted exactly the same.