The Prediction Profiler in JMP is a great way to interact with complicated statistical models. With the profiler you can:
-interactively generate predictions from your model as you change settings of individual factors,
-visualize interactions, like how the effect of one factor on your response can change at different levels of another factor,
-find optimal settings for your factors based on goals you have set for your response variables,
-and perform "what-if" analysis by simulating responses based on specified distributions for both factors and responses
If you have previously saved a model as a formula column, you can access the Prediction Profiler directly from the Graph menu. Otherwise, access the Prediction Profiler from the Red Triangle in a Model Report. Using the Tire Tread sample data, I've used Fit Model to generate a Response Surface Model for the variable "Hardness" based on several factors. The Prediction Profiler is showing a cross-section of the model. The prediction of Hardness is on the left, with a confidence interval, based on the current settings of silica, silane, and sulfur.
The curves for each factor shows the relationship between the response, Hardness, and each factor, at the current levels of the other two factors. For example, if we look at the plot with Silica, you see that the relationship between Hardness and Silica has curvature when Silane and Sulfur are at middle values, and that lower values of Silica appear to result in more hardness. As we change the value of Silica by dragging the red trace lines, the prediction of the model on the left changes.
With the Prediction profiler, we can also see interactions among our factors. For example, let's see how the relationship between Hardness and Silica changes at different levels of Silane. I'll drag Silane to the low and high levels, but focus your attention on the plot on the left of Hardness by Silica.
The Prediction Profiler helps us see this statistical interaction: at low levels of Silane,
higher
levels of Silica are associated with more Hardness. But, at high levels of Silane,
lower
levels of Silica are associated with more hardness.
In complicated models with interactions, it can be very difficult to manually find settings of our factors that give us the optimal response, which in this case is to achieve a Hardness of exactly 67.5. The Prediction Profiler can find the optimal settings for us. I'll select the Prediction Profiler Red triangle and select Optimization and Desirability then Desirability Functions. Desirability Functions are added to the Profiler, which show how desirable particular levels of our factors are if we are trying to match a Hardness target of 67.5.
To let JMP find the optimal factor settings, I'll select the Prediction Profiler Red triangle and select Optimization and Desirability then Maximize Desirability.
To achieve a hardness of 67.5, we can use the factor settings the Prediction Profiler has found for us, Silica at 0.42, Silane at 36, and Sulfur at 3.1.
Finally, the Prediction Profiler has a rich set of tools to perform "What-IF" analysis using Monte-Carlo Simulation. I'll select the Prediction Profiler Red triangle and select Simulator, to turn on the Simulation tools. Here I can set distributions for each of my factors if their values cannot be completely fixed, and when I click Simulate I can see the distribution of Hardness we can expect.
The Prediction Profiler in JMP is a powerful tool for understanding your models. To learn more, search Prediction Profiler in the JMP help.
