CRT Subcommand (TREE command)

The CRT subcommand sets parameters for the CRT method. For other tree growing methods, the subcommand is ignored, and a warning is issued.

Each keyword in the subcommand is followed by an equals sign (=) and the value for that keyword.

Example

TREE risk [o] BY income age creditscore
 /METHOD TYPE=CRT
 /CRT IMPURITY=TWOING MINIMPROVEMENT=.001.

IMPURITY Keyword

For categorical dependent variables, the IMPURITY keyword specifies the impurity measure used.

GINI. Gini is based on squared probabilities of membership for each category of the dependent variable. It reaches its minimum (zero) when all cases in a node fall into a single category. This is the default.

TWOING. Twoing groups categories of the dependent variable into the two best subclasses and computes the variance of the binary variable indicating the subclass to which each case belongs.

ORDEREDTWOING. A variant of twoing for ordinal dependent variables in which only adjacent classes can be combined. This method is available only for ordinal dependent variables only.

For scale dependent variables the least squared deviation (LSD) measure of impurity is used. It is computed as the within-node variance, adjusted for any frequency weights or influence values.

MINIMPROVEMENT Keyword

MINIMPROVEMENT specifies the minimum decrease in impurity for CRT trees. A node is not split if impurity would decrease less than the specified value.

• The default value is 0.0001.
• Specify a positive value. Large improvement values tend to result in smaller trees.