The Stepwise personality of the Fit Model platform features methods for stepwise regression techniques. The Stepwise personality computes estimates that are the same as those for models that are built in other least squares platforms, and it also facilitates variable selection.
Stepwise regression can use forward, backward, or a combination of the two for variable selection. The Stepwise personality includes options for each of these variable selection methods, as well as all possible regressions. In general, variable selection techniques are impacted by the order in which factors are added or removed from a model, the collinearity of factors, and the correlation of hypotheses tests that are used at each step to add or remove factors. The significance levels of the statistics for selected models violate the standard statistical assumptions because the model has been selected rather than tested within a fixed model. As with any model, the validity and value of the model should be based on the intended use of the model. A model for prediction should be validated on an independent data set. Factor settings that are determined from a model should be tested before widespread implementation. The book Subset Selection in Regression (Miller 1990) brings statistical understanding to model selection statistics.
Note: This chapter uses the term significance probability in a mechanical way to represent that the calculation would be valid in a fixed model.