To analyze a factorial design, the significant factor effects must be identified and separated from the insignificant effects and pure error. Once terms are selected, the model can be tested for significance on the ANOVA.
Some tips for analysts:
The Half-Normal plot is the default and recommended method to select effects for the model.
The Normal plot doesn’t usually provide more information than the half-normal plot.
The Pareto is best used to determine if the next largest term should be part of the model or left out of the model and pooled with the error estimates.
Numeric selection can be used to manually choose terms for the model or Automatic Model Selection algorithms can be used to select the terms for the model.
Use the Selection Tool to move from one method to another.
No matter which combination of tools are used, the best model selector is an informed analyst. Take moment and judge the model to make sure it makes sense from a scientific point of view. If it doesn’t make sense, look for problems with the data. Pay extra attention to the diagnostic plots if the analysis process is continued.