Constrain Response

[#1]

Hello everyone,

I'm stuck on a model what I first thought to be an easy one.. So I did a screen with 5 continuous factors and initially one response that I want to minimize. Later I needed to add a second response since I'm limited by inherent properties of the runs. The second response should not go over 6 while the first response should be minimal. However, I don't want to include my second response in the model creation but rather constrain the first response to yield a minimum in the desired range (second response <6). About half of my data is above the threshold of response 2 thus I don't want to exclude them.

Is there a way to do this in JMP?

Thanks and all the best,

FactorAlligator

[#2 · Accepted solution]

Hi @FactorAlligator,

You're right not to exclude data from the modeling step even if it doesn't match your final requirements. I don't know if the option I will share will exactly suit your needs, but here is a possibility :

You can create a model based on these 2 responses (make sure to check the option "Fit Separately" in the Fit Model platform) with the terms you specify.

In the red triangle of the Fit Group, click on Profiler (Profilers in Fit Group Reports (jmp.com)). From here, you will be able to specify for each response the target, limits and importance (in your example, you'll have to adjust the relative weight of your constraint response to be sure the optimum found by JMP will respect your constraint).

There is a lot of flexibility on how to set up your desirability function (Desirability Profiling and Optimization (jmp.com)), so we can imagine that you may set up the desirability profile of your response and "constraint response" like this :

This way, even if you're not interested in the exact values from response 2 (constraint), creating a desirability function for this response will enable to be sure that this constraint is taken into consideration in the Profiler optimum.

I hope this answer will help you,

[#3]

Hi @Victor_G,

thank you so much! This group profiler was the missing link and the solution you provided was exactly what I was looking for!

Best,

FactorAlligator

[#4]

Perfect, glad that it helped you !
There might be other solutions as well, but this option is quite straightforward, provided the second response is influenced by the same factors (so you are able to have a "good enough" model of the second response to be used in the Profiler).

All the best,

[#5]

This was my today's challenge. Thanks to your input, Victor, I solved it! Merci!