Interpret Prediction Profiler

There are several important points to note when interpreting a prediction profile:

• The importance of a factor can be assessed to some extent by the steepness of the prediction trace. If the model has curvature terms (such as squared terms), then the traces might be curved.

• If you change the value of a factor, the prediction trace for that factor is not affected, but the prediction traces of all the other factors can change. The Y response line must cross the intersection points of the prediction traces with their current value lines.

Note: If there are interaction effects or cross-product effects in the model, the prediction traces can shift their slope and curvature as you change current values of other terms. That is what interaction is all about. If there are no interaction effects, the traces change only in height, not slope or shape.

Figure 2.3 Changing One Factor from 0 to 0.75 

Prediction profiles are especially useful in multiple-response models to help judge which factor values can optimize a complex set of criteria.

Click a graph or drag the current value line right or left to change the factor’s current value. The response values change as shown by a horizontal reference line in the body of the graph. Double-click in an axis to bring up a window that changes its settings.