Methods and formulas for stepwise predictor selection for Fit Cox Model with Fixed Predictors only

Performs variable selection by adding or deleting predictors from the existing model based on the test of significance for the coefficient. In stepwise selection, the initial model is empty by default. Specifications for the analysis allow the addition of terms to the initial model and the addition of terms to every model. Stepwise is a combination of forward selection and backward elimination procedures. First, the procedure assesses whether to remove a term with the rules for backward elimination. If the procedure finds no terms to remove, then the procedure assesses whether to add a term with the rules for forward selection. Stepwise selection does not proceed if the initial model uses all of the degrees of freedom.

In forward selection, the initial model is empty or contains terms that are in every model. For every candidate term, Minitab Statistical Software calculates a the score test statistic and a corresponding p-value. If at least one candidate term has a p-value smaller than the value specified in Alpha to enter, then the term with the smallest p-value enters the model. With certain specifications for the analysis, additional terms enter the model in one step to maintain model hierarchy. Once added, a term never exits the model. The default forward selection procedure ends when none of the candidate terms have a p-value smaller than the value specified in Alpha to enter.

In backward elimination, the initial model contains all of the candidate terms. For every term in the model, Minitab Statistical Software calculates a Wald test statistic and a corresponding p-value. If at least one term has a p-value greater than the value specified in Alpha to remove, then the term with the largest p-value exits the model. With certain specifications for the analysis, additional terms exit the model in one step to maintain model hierarchy. Once removed, a term never re-enters the model. The default backward elimination procedure ends when none of the variables in the model have a p-value greater than the value specified in Alpha to remove. Backward elimination does not proceed if the initial model uses all of the degrees of freedom.