STEPWISE Subcommand (NOMREG command)
The STEPWISE subcommand gives you control of the statistical criteria when stepwise
methods are used to build a model. This subcommand is ignored if a
stepwise method is not specified on the MODEL subcommand.
RULE(keyword). Rule for entering
or removing terms in stepwise methods. The default SINGLE indicates that only one effect can
be entered or removed at a time, provided that the hierarchy requirement
is satisfied for all effects in the model. SFACTOR indicates that only one effect can be entered
or removed at a time, provided that the hierarchy requirement is satisfied
for all factor-only effects in the model. CONTAINMENT indicates that only one effect can be entered
or removed at a time, provided that the containment requirement is
satisfied for all effects in the model. NONE indicates that only one effect can be entered or removed at a time,
where neither the hierarchy nor the containment requirement need be
satisfied for any effects in the model.
MINEFFECT(n). Minimum number
of effects in final model. The default is 0. The intercept,
if any, is not counted among the effects. This criterion is ignored
unless one of the stepwise methods BACKWARD or BSTEP is specified.
MAXEFFECT(n). Maximum number
of effects in final model. The default value is the total
number of effects specified or implied on the NOMREG command. The intercept, if any, is not counted
among the effects. This criterion is ignored unless one of the stepwise
methods FORWARD or FSTEP is specified.
ENTRYMETHOD (keyword). Method for
entering terms in stepwise methods. The default LR indicates that the likelihood ratio test
is used to determine whether a term is entered into the model. SCORE indicates that the score test is used.
This criterion is ignored unless one of the stepwise methods FORWARD, BSTEP, or FSTEP is specified.
REMOVALMETHOD (keyword). Method for
removing terms in stepwise methods. The default LR indicates that the likelihood ratio test
is used to determine whether a term is entered into the model. WALD indicates that the Wald test is used.
This criterion is ignored unless one of the stepwise methods BACKWARD, BSTEP, or FSTEP is specified.
PIN(a). Probability
of the likelihood-ratio statistic for variable entry. The
default is 0.05. The larger the specified probability, the easier
it is for a variable to enter the model. This criterion is ignored
unless one of the stepwise methods FORWARD, BSTEP, or FSTEP is specified.
POUT(a). Probability
of the likelihood-ratio statistic for variable removal. The default is 0.1. The larger the specified probability, the easier
it is for a variable to remain in the model. This criterion is ignored
unless one of the stepwise methods BACKWARD, BSTEP, or FSTEP is specified.
The hierarchy requirement stipulates that among
the effects specified or implied on the MODEL subcommand, for any effect to be in a model, all lower-order effects
that are part of the former effect must also be in the model. For
example, if A, X, and A*X are specified, then for A*X to be in a model,
the effects A and X must also be in the model.
The containment requirement stipulates that among
the effects specified or implied on the MODEL subcommand, for any effect to be in the model, all effects contained
in the former effect must also be in the model. For any two effects
F and F’, F is contained in F’ if:
- Both effects F and F’ involve the same covariate effect, if any. (Note that effects A*X and A*X*X are not considered to involve the same covariate effect because the first involves covariate effect X and the second involves covariate effect X**2.)
- F’ consists of more factors than F.
- All factors in F also appear in F’.
The following table illustrates how the hierarchy
and containment requirements relate to the RULE options. Each row of the table gives a different
set of effects specified on the MODEL subcommand. The columns correspond to the RULE options SINGLE, SFACTOR, and CONTAINMENT. The cells contain the order
in which effects must occur in the model. For example, unless otherwise
noted, all effects numbered 1 must be in the model for any effects
numbered 2 to be in the model.
| Effects | SINGLE | SFACTOR | CONTAINMENT |
|---|---|---|---|
| A, B, A*B | 1. A, B 2. A*B |
1. A, B 2. A*B |
1. A, B 2. A*B |
| X, X**2, X**3 | 1. X 2. X**2 3. X**3 |
Effects can occur in the model in any order. | Effects can occur in the model in any order. |
| A, X, X(A) | 1. A, X 2. X(A) |
Effects can occur in the model in any order. | 1. X 2. X(A) Effect A can occur in the model in any order. |
| A, X, X**2(A) | 1. A, X 2. X**2(A) |
Effects can occur in the model in any order. | Effects can occur in the model in any order. |