The HPSPLIT Procedure

MODEL Statement

  • MODEL response <(response-option)> = variable <variable...>;

The MODEL statement causes PROC HPSPLIT to create a tree model by using response as the response variable and variable as a predictor. By default, variable is treated as a continuous predictor if it is a numeric variable, or as a categorical variable if the variable also appears in the CLASS statement.

Note: Specifying a character variable in a MODEL statement without previously declaring it in a CLASS statement results in an error.

You can specify the following response-option:

EVENT='category'

specifies the event level for a binary categorical response variable. PROC HPSPLIT associates this level with the event of interest (sometimes referred to as the positive outcome) for the purpose of computing sensitivity, specificity, and area under the curve (AUC) and creating receiver operating characteristic (ROC) curves. You can specify the value (formatted if a format is applied) of the event category in quotation marks.

The default level is the first level of the response variable as specified by the ORDER= option in the CLASS statement.

Note: The EVENT= option has no effect in the case of categorical response variables that have more than two levels, and it has no effect in the case of continuous response variables.