Industrial Statistics with Minitab

23.6 Selection of the Best Equation: Stepwise

Stat > Regression > Stepwise

With the previously used option (Best Subsets), it is possible to consider up to 31 explanatory variables as candidates to enter into the model. Although a large number of variables (from 20 or 25) implies a high computing time, this option is usually sufficient in most cases.

If the number of variables is greater than 31, or the computing time is very high when using less than 31 variables, an alternative is to use the Stepwise method, which is known to be very fast regardless of the number of explanatory variables.

23.6.1 Methods

Default values:

Stepwise: The most common method. The variables are entered into the model or removed from the model at each step. It is possible to choose a set of variables to be in the model when the process starts (Predictors in initial model). By default, the initial model does not include any variable.

Forward selection: Once the variables are included in the model they cannot be removed.

Backward elimination: The variables that have been included in the model can be removed (all variables initially included in the model), but once a variable is removed, it never enters again.

Use all default options to obtain:

You must enter the requested subcommand when obtaining ...