Practical Data Analysis with JMP®

Evaluating Alternative Models

With multiple regression analysis we often want to compare two or more competing models. In our concrete example, for instance, we earlier estimated a model with three significant factors, reasonably well-behaved residuals, and which accounted for about 72% of the observed variation in compressive strength. Perhaps a four-variable model would do better?

The process of comparing models can easily devolve into a chaotic flurry of trial-and-error attempts. When estimating several models for a single Y variable, take these points into consideration:

• Do all of the independent factors make logical sense in the model? Are there solid theoretical reasons to include each factor? If not, why include nonsense factors?

• Does ...