Surrogates

The differences between the CHAID and CRT models are due, in part, to the use of surrogates in the CRT model. The surrogates table indicates how surrogates were used in the model.

Figure 1. Surrogates table
Surrogates table
  • At the root node (node 0), the best independent (predictor) variable is number of credit cards.
  • For any cases with missing values for number of credit cards, car loans is used as the surrogate predictor, since this variable has a fairly high association (0.643) with number of credit cards.
  • If a case also has a missing value for car loans, then age is used as the surrogate (although it has a fairly low association value of only 0.004).
  • Age is also used as a surrogate for income level at nodes 1 and 5.

Next