What inspired this Wish List Request? A recent Customer JMP Champion (Derrick Carter-House - Sr. Scientist Clear Labs - Derrick.carter-house@clearlabs.com) asked what is the difference between the Profiler Prediction Profiler compared to the Profiler Variable Importance Marginal Model Plot. The customer found the on-line hope to be too technical and unhelpful and reached out to his SE Support/Global Enablement for an understandable answer.
Please describe the new or improved feature or solution you would like to see? Here is the current JMP Help section on Profiler Variable Importance Marginal Model Plots with the technical description only of the methodology mechanics. Current: JMP 17.1 Help - Assess Variable Importance Marginal Model Plots
What is missing is a clear description of how different than the other Profilers and when to use the Margin Model Plots. See the following which was taken from Jian Cao's (@jiancao) research and clear explanation back to the customer of the difference and how utilized.
The answer to the question of how JMP Profiler’s Marginal Model Plot is different from the Prediction Profile is that the purposes of these plots are different. Prediction Profilers show the relationship between fitted Y (i.e., OLS) and observed values of factors to assess model adequacy. However, the Marginal Model Plots shows the relative importance of factors based on the smoothing of the simulated values. They rather provide a visual display of factor “importance” by fitting as smoothers to the simulated values of predicted and factors. For example, the Marginal Model Plot was run on a model built in the JMP Help data set Car Physical Data.jmp data:
- Displacement has the tightest confidence intervals indicating it being the most important (we already know it accounts for 76% of the variability in precited Y).
- If you select the option to add observed data points to the above Marginal Model Plots you can see how actual data points will be off compared to the smoothed simulated data.
Additional: If interested, Brady Brady (@brady_brady ), another technical expert from Global Enablement, also created a Marginal Model Plot JMP Add-In that uses a different method for the other purpose of assessing model accuracy. (See Marginal Model Plots - JMP User Community)
See Car Physical Data attachment with Marginal Model Plot script.
What would the value be to you and/or other users if this new feature is provided? Other users would have more interest in Profiler Variable Importance and Marginal Model Plots to better explore their models if there was stronger description in the help documentation.
Is this an external customer request? Customer JMP Champion (Derrick Carter-House - Sr. Scientist Clear Labs - Derrick.carter-house@clearlabs.com). Feel free to contact.
Is there a competitive software package that has this feature or solution that we should be aware of? Unknown, perhaps checked with Product Management.
Will this impact a renewal, expansion, or adoption at a customer, and does this put an account at risk if we do not address this? Small impact in encouraging continual and deeper JMP use to promote new and renewal sales growth.
