Robust process setting with a categorical noise effect

[#1]

With profiler it is not possible to use a categorical noise effect? Is there another way to specify robust process settings?

[#2]

Effects or Responses

[#3]

Find robust process settings for a categorical noise factor on a continuous response

[#4 · Accepted solution]

Hi @frankderuyck,

From what I understand from your situation, there may be several options available :

1. You can consider your categorical factor as a random effect (or noise), as you're interested on the variance caused by this factor, not by the change on the mean response.
2. You can also use the Simulator, to add random noise to a categorical factor and assess how this may change the response distribution.

For the suggestion 1), I used the sample dataset "DOE Example 1" in the JMP sample index (in DoE). I changed the design role property of "Operator" to "Random Block", and run a model with the following effects :

I end up with a Profiler that takes into account the variability introduced by the random effect "Operator" :

If using the profiler with Operator as a categorical fixed effect (or if set up as random effect, I can click on the red triangle next to the profiler, and make "Operator" factor appears by clicking on "Conditional Predictions"), I can use the Simulator, specify the type of noise (here I choose "Random" with a probability of 50% for each operator) and then click on Simulate :

The two approaches are conceptually different (a random effect suppose that these 2 operators are part of a bigger, unknown population which may also impact the variance of the results, whereas the Simulator uses only these 2 possible options to compute the results distribution) and lead to different results, so depending on your topic you may find one more relevant than the other.

I hope this will help you,

[#5]

Very useful Victor, thanks!