Regardless of the split selection criterion, however, in each node the variable that is most strongly associated with the response variable (i.e., that produces the highest impurity reduction or the lowest p-value) is selected for the next split. In splitting variables with more than two categories, that offer more than one possible cutpoint, the optimal cutpoint is also selected with respect to this criterion. In our example, the optimal cutpoint identified within the range of the numeric predictor variable alcohol_per_month is between the values 1 and 2, because subjects who drank alcohol in one or less occasions have a lower frequency of “yes” answers than those who drank alcohol in 2 or more occasions.
