JMP response out of defined bounds

[#1]

Hi everyone, I wanted to let you know that I am using the 'fit model' option in JMP with two response parameters out of which one is to maximize and other is to minimize, both are modelled together. On the first response (maximise) only lower bound (+ve number) is set whereas for the 2nd response (minimize), both the bounds, lower and upper (both bounds +ve numbers) are applied.
However, I have observed that using the 'Maximize Desirability' is finding the solution for second response which is out of these  above bounds i.e. the resulting response is smaller than the lower bound. Additionally, for this response, the resulting lower value of its confidence interval comprise -ve value as well and upper value of this interval is slightly higher than the defined bounds post 'Maximize Desirability' operation.

Is it possible that JMP can provide these outputs (out of bounds) even after defining the above constraints or am I missing something? Any clues? Thanks in advance.

[#2]

Are you referring to the Response Limits column property? It defines the desirability function for the response. It does not limit the range of the response value in prediction. How is this property defined for the two responses?

Linear regression assumes that the response can vary from negative to positive infinity. You might be able to use a transform function on the response. The logit transform (log odds) can work with a range of [0,1] but provide a response for regression that meets the assumptions.

[#3]

Hi Mark, thank you very much for your response. Yes, you are correct, the response bounds are defined in Response Limit column property and for both the responses, the properties defined are continuous and numeric. The description besides Response Limit Box refers (below) it as bounds on the response's range of acceptability, so I assumed JMP would find solutions considering the response bounds as constraints.

ProbitStarfish1_0-1664875065559.png

Is there way to define the constraints on the response values when using the 'Fit Model' to obtain the solution that results in predicted response to be within the constraints so the output can be more sensible?

[#4]

Is the issue that you have that the linear regression model predicts a response beyond the 'bounds' that you set with the Response Limits column property? The upper bound for Maximize says that desirability attains a level of 1 for this response or greater. It does not limit it such that it cannot go higher when maximizing the desirability, which might be necessary in order to optimize another response.

What is the purpose of the lower bound for the response to be minimized?

What is the nature of the responses?

[#5]

Hello, thank you very much for your answer earlier on this. Would you be able to throw in your opinion on how to define the bounds that can prevent the JMP finding the -ve values of response or find the solutions that can limit the response within the defined bounds.

Thank you very much.

[#6]

Such bounds do not exist in JMP at this time. You might be able to get what you want by defining a second data column that is identical to the original column but defining the opposite Response Limits column property. For example, I have response Y1 with a Response Limits property defined with a Maximize goal, a lower limit of 0, and an upper limit of 1. I could create another response, Y2, with the same prediction formula but define the Response Limits property with a Minimize goal, a lower limit of 1, and an upper limit of 2 or higher. Use both Y1 and Y2 with Graph > Profiler. The desirability should stop at 1.