How to limit a response prediction to a threshold in a model?

[#1]

Hi!

I want to model particle size, which is a variable that can't be below 0. However, some setups are being predicted to be negative (see picture below). So my question is how can I limit the prediction threshold to be above zero? Is there any truncated approach in JMP?

[#2 · Accepted solution]

Hi @PValueEnemy 

It may be difficult to help you only based on this residuals plot and your topic description, but here are some suggestions from similar topic :Fit Model - Limiting Model Response to Only Positive Values : You can use response transformation to avoid negative predictions.

• For particules size, since you might have very broad sizes range, a log transformation could be helpful.
• You can also in the Fit Model report check the Box-Cox Y Transformation to see if it helps for the predicted values and the respect of the regression assumptions : Regression Model Assumptions | Introduction to Statistics | JMP. The plot you shown here doesn't seem to show that errors are normally distributed for example.
• You could also use Generalized Regression models with specific set distributions, like LogNormal.

You can use the platform Distributions to check the most probable data distribution(s) of your response(s) and use this information in your response transformation or in setting an appropriate distribution in your Generalized Regression model.

Hope this first discussion starter might help,

[#3]

Thanks, @Victor_G 

Your reply helped a lot. I suppressed many information to reduce the question, but you were right where I needed