Adding, Removing, and Moving Predictor Attributes: The Goal of Model Building

How To... > Work with Statistical Objects > Build a Linear Model > Multiple Regression > Adding, Removing, and Moving Predictor Attributes: The Goal of Model Building

You can drag additional attributes to the table of attributes. Notice that as you drag, you can decide where in the table an additional attribute goes, as signified by the location of the thick black line. (You can drag more than one attribute at a time to the table.)

You can remove an attribute by selecting its line in the table of attributes and then choosing Model | Remove Attribute.

You can move an attribute within the table by dragging it to a new location. Notice that the portion of the ribbon chart corresponding to an attribute turns red when you select the attribute. Moving the attribute changes its place in the table, the regression equation, and the ribbon chart. Notice that moving things around does not change the coefficient for any of the attributes or the total R2. It does, however, change the sequential contribution.

You might well ask, “What is the goal of all this adding, removing, and moving of attributes? How do I know when I’m done?” Building a multiple regression model is a bit of an art. There’s no single right answer. To some extent, you’re trying to account for as much of the variation as possible. But you also want the model to be as simple as possible, and you want the attributes to make sense as you interpret the model.

Think again about the Airplanes model. Doesn’t it seem likely that fuelgph would be better at explaining cost than any of the other three possible predictor attributes? In fact, the model with fuelgph as the only predictor accounts for more of the variation than the model with all of the other three! Simple linear regression turns out to be simpler and better.